The openness profiles · snapshot Nov 15, 2026

How open is it?

What did the publisher actually share? Look inside OLMo or gpt-oss, or choose a tier to explore releases with similar freedoms.

← Higher release tierTiers describe the freedoms that come with a release.Lower release tier →
Unscored reference entries, outside the spectrum.
AAccessPPermissionDDeploymentTTransparencyRReproducibilityGGovernanceEach scored 0–3How scoring works

The six questions in the book →

AAccess
Are weights and inference artifacts downloadable?
PPermission
Can users use, modify, and redistribute the artifacts for their purpose?
DDeployment
Can users run the weights and runtime independently on chosen hardware?
TTransparency
Are training data, data processing, post-training, alignment, and evaluation documented?
RReproducibility
Can a skilled group build a substantially equivalent system?
GGovernance & Participation
Can a person participate in developing this model — contributing data, code, evaluation, or direction — without being employed by or affiliated with the publisher? Is there a genuine community with an open development process, or is it a one-way corporate release?
  • 0 No
  • 1 Limited / partial
  • 2 Yes, with caveats
  • 3 Strong / exemplary

A named release or explicitly listed group with equivalent terms. Mixed-license publisher/family summaries are reference overviews, not scored artifacts. Training toolkits and method implementations are identified separately from pretrained models; their access/rebuild scores describe code and method, not undisclosed pretrained weights.

How the book assigns tiers →

Tier5
Reproducible open release
A, P, D, T and R all score 3. Published resources support rebuilding the assessed artifact; this tier alone is not an OSAID certification.
Tier4
Documented research release
A=3, P>=2, D=3, T>=2 and R>=2, but the release does not meet all Tier 5 conditions. Permission and reconstruction caveats remain visible.
Tier3
Practical open-weight
A=3, P>=2 and D=3, but Transparency or Reproducibility is below 2.
Tier2
Restricted or deployment-limited
A=3 and D>=2, but Permission is below 2 or Deployment is below 3.
Tier1
Access-limited
Some runnable artifacts are public, but Access is below 3 or Deployment is below 2.
Tier0
Proprietary service
Access and Deployment both score 0: no independently runnable release is available.

Assignment rules, first match

  1. Tier 0If A = 0 and D = 0
  2. Tier 1Otherwise, if A < 3 or D < 2
  3. Tier 2Otherwise, if P < 2 or D < 3
  4. Tier 5Otherwise, if A = P = D = T = R = 3
  5. Tier 4Otherwise, if T ≥ 2 and R ≥ 2
  6. Tier 3Otherwise

Tiers are derived deterministically from A, P, D, T and R using the ordered rules below. G is an independent dimension. Select the highest matching release tier after the access/deployment gates. Scores reflect documented author judgment; no tier alone certifies compliance with the Open Source AI Definition.

Scored profiles carry author-assigned dimensions and a derived tier. Family overviews have neither. Within each tier, profiles are alphabetical; no hidden sum ranks them.

91 profiles

Tier5

Reproducible open release

A, P, D, T and R all score 3. Published resources support rebuilding the assessed artifact; this tier alone is not an OSAID certification.

8 profiles
Tier5AutoGluon (tabular, multimodal and time-series AutoML)Amazon Web Services and the AutoGluon community Classic MLAutoML with the search space exposed rather than hidden behind a service.APDTRGApache 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
3/3 Strong / exemplary
GGovernance
3/3 Genuine community governance

Tier rationale

A=P=D=T=R=3. A=3 P=3 D=3 T=3 R=3. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Training toolkit
Assessment recorded

Apache 2.0

  • Some bundled foundation-model backbones carry their own terms

Why it matters

The contrast with proprietary AutoML services is that the automation stays auditable and forkable.

Tier5EleutherAI Pythia scaling suiteEleutherAI Text LLMsThe cleanest true reproducibility case: same data, same order, every checkpoint published.APDTRGApache 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
3/3 Strong / exemplary
GGovernance
3/3 Genuine community governance

Tier rationale

A=P=D=T=R=3. A=3 P=3 D=3 T=3 R=3. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0

Why it matters

The benchmark case for what fully open means, and the reason the book can show that reproducibility is a design choice rather than an accident.

Tier5H2O-3 AutoMLH2O.ai Classic MLAutomated model selection, fully open.APDTRGApache 2.0 for open-source H2O-3; H2O-3 Secure (AutoDoc, SSO and RBAC, MOJO scoring into Hadoop and Spark, SLAs) is a closed commercial offeringArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
3/3 Strong / exemplary
GGovernance
2/3 Real but bounded participation

Tier rationale

A=P=D=T=R=3. A=3 P=3 D=3 T=3 R=3. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Training toolkit
Assessment recorded

Apache 2.0 for open-source H2O-3; H2O-3 Secure (AutoDoc, SSO and RBAC, MOJO scoring into Hadoop and Spark, SLAs) is a closed commercial offering

  • Several enterprise features are closed in H2O-3 Secure

Why it matters

AutoML is the machine-learning equivalent of a published recipe, and the enterprise tier shows how openness gets carved back into features.

Tier5KataGo (self-play Go engine)KataGo project (David J Wu and contributors; earlier runs supported by Jane Street) Decision & gamingThe closest thing in the batch to fully open and reproducible: code, data, weights and method.APDTRGMIT-style licence (custom text with an additional disclaimer); network weights published by the project for free useArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
3/3 Strong / exemplary
GGovernance
3/3 Genuine community governance

Tier rationale

A=P=D=T=R=3. A=3 P=3 D=3 T=3 R=3. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Method implementation
Assessment recorded

MIT-style licence (custom text with an additional disclaimer); network weights published by the project for free use

  • Bundled third-party libraries under their own licences

Why it matters

The cleanest positive case, a frontier-adjacent game-playing system where nothing material is withheld.

Tier5LightGBMMicrosoft Classic MLMIT-licensed gradient boosting with no attachments.APDTRGMITArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
3/3 Strong / exemplary
GGovernance
3/3 Genuine community governance

Tier rationale

A=P=D=T=R=3. A=3 P=3 D=3 T=3 R=3. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Training toolkit
Assessment recorded

MIT

Why it matters

Confirms the classical-ML openness baseline across independent implementations rather than a single project.

Tier5OLMo 2 / OLMo 3 / OLMo 3.1Allen Institute for AI (Ai2) Text LLMsOpen research system: the reference case for end-to-end openness.APDTRGApache 2.0 on released model weights and project code; training and fine-tuning resources retain their stated dataset termsArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
3/3 Strong / exemplary
GGovernance
2/3 Real but bounded participation

Tier rationale

A=P=D=T=R=3. A=3 P=3 D=3 T=3 R=3. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0 on released model weights and project code; training and fine-tuning resources retain their stated dataset terms

Why it matters

OLMo provides a substantial published model-development flow, enabling investigation and reconstruction attempts beyond running a finished checkpoint.

Tier5scikit-learnscikit-learn developers Classic MLThe default Python ML library, permissively licensed and fully inspectable.APDTRGBSD 3-ClauseArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
3/3 Strong / exemplary
GGovernance
3/3 Genuine community governance

Tier rationale

A=P=D=T=R=3. A=3 P=3 D=3 T=3 R=3. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Training toolkit
Assessment recorded

BSD 3-Clause

Why it matters

A decade of accumulated practice sits in a repository anyone can read and modify, the opposite of an API-only frontier model.

Tier5XGBoostDMLC (Apache Software Foundation project) Classic MLFully open by construction: the artifact is the method.APDTRGApache 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
3/3 Strong / exemplary
GGovernance
3/3 Genuine community governance

Tier rationale

A=P=D=T=R=3. A=3 P=3 D=3 T=3 R=3. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Training toolkit
Assessment recorded

Apache 2.0

Why it matters

The baseline the open-model debate is measured against; in classical ML, open is the default rather than a concession.

Tier4

Documented research release

A=3, P>=2, D=3, T>=2 and R>=2, but the release does not meet all Tier 5 conditions. Permission and reconstruction caveats remain visible.

13 profiles
Tier4AlphaFold 2 and AlphaFold-MultimerGoogle DeepMind ScienceDownloadable code, parameters and database, with share-alike databases and no training code.APDTRGCode Apache 2.0; parameters CC BY 4.0; mirrored databases mixed (BFD and UniRef30 CC BY-SA 4.0, MGnify and PDB CC0); AlphaFold DB CC BY 4.0Artifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
2/3 Yes, with caveats
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
1/3 Cosmetic participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=2 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Code Apache 2.0; parameters CC BY 4.0; mirrored databases mixed (BFD and UniRef30 CC BY-SA 4.0, MGnify and PDB CC0); AlphaFold DB CC BY 4.0

  • Share-alike obligations on BFD and UniRef30 mirrors
  • Attribution required for parameters and AFDB entries

Why it matters

The reference case for open in science: nearly everything needed to use and audit the system is downloadable, yet the training pipeline stays closed.

Tier4Apertus v1.5 70BSwiss AI Initiative (EPFL, ETH Zurich, CSCS) Text LLMsExtensive reconstruction resources alongside separately accepted download and redistribution obligations.APDTRGApache License 2.0 on released weights and project code; training-data inputs retain their own termsArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
2/3 Yes, with caveats
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
3/3 Strong / exemplary
GGovernance
2/3 Real but bounded participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=2 D=3 T=3 R=3. G is displayed separately.

Scope and terms

Assessment scope
The v1.5 70B release and its July 14, 2026 AUP. Older Apertus generations and mini distills are excluded.
Artifact kind
Trained model
Assessment recorded
Selected evidence checked

Apache License 2.0 on released weights and project code; training-data inputs retain their own terms

  • Apache notice requirements and trademark reservations apply; input datasets retain their original licenses.

Separate access agreements

  • The publisher's Hugging Face gate requires separate assent to the Apertus v1.5 AUP.
  • The AUP requires indemnification of ETH Zurich and EPFL and express downstream assent to the AUP in force at redistribution, with the obligation passed onward.

Why it matters

A release can disclose substantial training evidence while its acquisition contract still adds material obligations. Artifact-license approval and contractual permission are separate questions.

Tier4BigCode StarCoder2BigCode project (Hugging Face, ServiceNow, Red Hat and others) CodeCode model whose training data is genuinely open, including the opt-out mechanism.APDTRGBigCode OpenRAIL-MArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
2/3 Yes, with caveats
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
2/3 Yes, with caveats
GGovernance
3/3 Genuine community governance

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=2 D=3 T=3 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

BigCode OpenRAIL-M

  • use-based restrictions with permitted-use and prohibited-use lists
  • commercial use allowed but conditioned on use restrictions

Why it matters

Published code-training resources and documented opt-out handling make provenance inspectable. Availability alone does not establish that every included third-party item is free of legal claims.

Tier4DeepSeek (V3.2, V4 Pro, V4 Flash, V4.1 Flash)DeepSeek Text LLMsPermissive and influential open-weight family with substantial technical documentation and unresolved provenance debates.APDTRGMIT for the model weights and the accompanying code (deepseek-recipe, DeepInfer)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
1/3 Cosmetic participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=3 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the listed MIT-licensed V3.2, V4 Pro, V4 Flash and V4.1 Flash releases. Original V3 and older DeepSeek-LLM custom-license releases are excluded.
Artifact kind
Trained model
Assessment recorded

MIT for the model weights and the accompanying code (deepseek-recipe, DeepInfer)

  • Original V3 and older DeepSeek-LLM releases used custom model terms; DeepSeek-R1 adopted MIT. Check each derivative separately.

Why it matters

DeepSeek combines strong practical openness, permissive terms, efficient training claims, and controversy over possible distillation - capability, license, provenance, and allegations must stay separate.

Tier4DreamerV3Danijar Hafner (independent reimplementation; method from DeepMind research) World modelsOpen method, closed model: MIT code with no weights to open.APDTRGMITArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
2/3 Real but bounded participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=3 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Method implementation
Assessment recorded

MIT

  • The public repo is the author's reimplementation, explicitly unrelated to Google or DeepMind
  • No pretrained weights released by any party

Why it matters

Stretches the rubric honestly: Access measures artifacts that exist, and for decision models the artifact is often the method itself — a different shape of openness than a weight release.

Tier4ESM-2 and ESMFold (Meta FAIR protein language models)Meta Fundamental AI Research ScienceMIT code and MIT weights over public sequence data: the practical open-weight baseline for biology.APDTRGMIT for source code and released checkpoints; ESM Metagenomic Atlas data under CC BY 4.0 with Meta Open Source TermsArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
1/3 Cosmetic participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=3 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT for source code and released checkpoints; ESM Metagenomic Atlas data under CC BY 4.0 with Meta Open Source Terms

  • Meta Open Source Terms apply to Atlas data use
  • Verify terms per checkpoint

Why it matters

Shows the protein-modelling openness ceiling that AlphaFold 3 weight terms fall short of, and the cleanest MIT-plus-public-data example in the batch.

Tier4ESM3 open release (esm3-sm-open-v1, 1.4B)EvolutionaryScale, now Chan Zuckerberg Biohub ScienceA genuinely open small checkpoint sitting beside a closed flagship.APDTRGMIT for code and the open checkpoint (Biohub/esm LICENSE.md and ESM3 README); the 2024 release shipped a non-commercial Community License requiring acceptance on the HubArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
2/3 Real but bounded participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=3 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT for code and the open checkpoint (Biohub/esm LICENSE.md and ESM3 README); the 2024 release shipped a non-commercial Community License requiring acceptance on the Hub

  • Earlier snapshots were gated behind non-commercial terms; check the exact release you cite

Why it matters

Illustrates the common pattern where openness is granted for a reduced model while the capable model stays behind an API.

Tier4Hugging Face SmolLM (SmolLM2, SmolLM3)Hugging Face Text LLMsSmall model with an unusually complete public engineering blueprint.APDTRGApache 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
2/3 Real but bounded participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=3 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0

Why it matters

Demonstrates that meaningful reproducibility at small scale is achievable without publishing every pretraining byte.

Tier4Leela Chess Zero (Lc0)Leela Chess Zero community Decision & gamingOpen chess self-play at scale, with copyleft code and community-run training.APDTRGGNU GPL v3 for the engine, with an additional GPL section 7 permission for CUDA-linked builds; network weights distributed freely by the projectArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
2/3 Yes, with caveats
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
3/3 Genuine community governance

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=2 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Method implementation
Assessment recorded

GNU GPL v3 for the engine, with an additional GPL section 7 permission for CUDA-linked builds; network weights distributed freely by the project

  • Copyleft obligations on redistributed engine code

Why it matters

A decade-long, still-running open training run, which is the strongest evidence that decision systems can stay open in practice.

Tier4LeRobot (library, dataset format and policies)Hugging Face RoboticsAn open ecosystem layer: formats, datasets, training code and evaluation in one permissive repository.APDTRGApache 2.0, with parts derived from Diffusion Policy under MITArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
3/3 Genuine community governance

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=3 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Training toolkit
Assessment recorded

Apache 2.0, with parts derived from Diffusion Policy under MIT

  • Individual Hub datasets and policies carry their own licences; check per repository

Why it matters

Shows that openness in robotics can be delivered as shared infrastructure rather than as a single open model.

Tier4Open MuZero and AlphaZero implementations (mctx, alpha-zero-general, pgx)Google DeepMind for mctx; open-source community for the rest Decision & gamingDeepMind's game-playing method survives without DeepMind's weights, in permissively licensed code.APDTRGApache 2.0 for DeepMind mctx and pgx; MIT for alpha-zero-general and comparable projectsArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
3/3 Strong / exemplary
GGovernance
2/3 Real but bounded participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=3 D=3 T=2 R=3. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Method implementation
Assessment recorded

Apache 2.0 for DeepMind mctx and pgx; MIT for alpha-zero-general and comparable projects

  • No official DeepMind MuZero weights exist; these are independent implementations

Why it matters

When the lab withholds weights, the method can still be reproduced, so openness can come from the field rather than the publisher.

Tier4Segment Anything / SAM 2Meta AI (FAIR) VisionThe rare vision release where data, code, and weights all shipped.APDTRGApache 2.0 (code, checkpoints, SAM 2.1 included)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
1/3 Cosmetic participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=3 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0 (code, checkpoints, SAM 2.1 included)

  • SA-1B dataset under a separate Research License (non-commercial)

Why it matters

Published segmentation data and code support a reconstruction attempt. The data has its own terms, and the assessed transparency and reproducibility gaps remain visible; an Apache artifact license alone does not earn Tier 5.

Tier4TinyLlama 1.1BTinyLlama project (SZ LLMSphere) Small language modelsA small model trained entirely in public.APDTRGApache 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
3/3 Strong / exemplary
RReproducibility
2/3 Yes, with caveats
GGovernance
2/3 Real but bounded participation

Tier rationale

A=3, P>=2, D=3, T>=2 and R>=2; at least one dimension prevents Tier 5. A=3 P=3 D=3 T=3 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0

Why it matters

Proof that the Chinchilla-optimal small-model recipe can be executed end-to-end on open data — the reproducibility ceiling most frontier releases never approach.

Tier3

Practical open-weight

A=3, P>=2 and D=3, but Transparency or Reproducibility is below 2.

30 profiles
Tier3Alibaba GTE embeddings (gte-Qwen2, gte-base-en-v1.5, gte-modernbert-base, gte-multilingual-base)Alibaba Cloud (Alibaba-NLP) EmbeddingsApache-2.0 embedding family spanning Qwen2 and ModernBERT bases.APDTRGApache 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0

Why it matters

Completes the permissive-licence pattern across three independent embedding vendors.

Tier3Ant Group Ling (Ling-2.6-flash, Ling-2.6-1T)Ant Group (inclusionAI) Text LLMsPermissively licensed industrial open weights with a published training-content summary.APDTRGMIT (HF license:mit tag on Ling-2.6-flash, Ling-2.6-1T and their fp8/int4 variants)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT (HF license:mit tag on Ling-2.6-flash, Ling-2.6-1T and their fp8/int4 variants)

Why it matters

A useful reference point for how a large non-Western lab describes its training data without opening the corpus.

Tier3BAAI BGE embeddings (bge-m3, bge-small and large v1.5)BAAI EmbeddingsPermissively licensed embedding family that is the de facto default for retrieval.APDTRGMIT for bge-m3 and the v1.5 encoders; Apache 2.0 for bge-reranker-v2-m3 - verify per checkpointArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT for bge-m3 and the v1.5 encoders; Apache 2.0 for bge-reranker-v2-m3 - verify per checkpoint

Why it matters

Embeddings carry the same openness questions as generation models, but the licence answer here is unusually clean.

Tier3Cohere North-Mini-Code 1.0 and 2026 Apache-2.0 releasesCohere (Cohere Labs) CodeSame vendor, different license: Cohere 2026 releases are permissive Apache 2.0.APDTRGApache 2.0 (North-Mini-Code-1.0, cohere-transcribe-03-2026, cohere-transcribe-arabic-07-2026)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0 (North-Mini-Code-1.0, cohere-transcribe-03-2026, cohere-transcribe-arabic-07-2026)

Why it matters

Isolates the license variable - the openness of a release is decided by the terms attached, not by the brand.

Tier3DINOv2Meta AI (FAIR) VisionFully permissive self-supervised vision backbone.APDTRGApache 2.0 (code and released weights)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0 (code and released weights)

  • X-ray DINO weights: FAIR Noncommercial Research License
  • Cell-DINO code: CC-BY-NC

Why it matters

The earlier CC-BY-NC framing aged out — DINOv2 shows a lab re-releasing its vision line under a true open-source license.

Tier3Google Gemma 4Google DeepMind Text LLMsA major lab family that moved to a standard license without opening the training system.APDTRGApache 2.0 for the Gemma 4 weights, a change from the custom Gemma terms used for earlier generationsArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded
Selected evidence checked

Apache 2.0 for the Gemma 4 weights, a change from the custom Gemma terms used for earlier generations

  • Earlier Gemma releases (1 to 3) shipped under a custom Gemma license with a prohibited-use policy and should be checked separately
  • The license does not cover Google trademarks, Google's API services, or the underlying training data

Why it matters

Gemma 4 is the clearest case of a standard software license on weights coexisting with an unchanged training system, and of a corporate lab answering the license debate without answering the reproducibility one.

Tier3GPT-2OpenAI Small language modelsThe original open-weight model, and the original missing-dataset problem.APDTRGModified MIT License (adds an output-ownership paragraph; not verbatim OSI-approved MIT text)Artifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Modified MIT License (adds an output-ownership paragraph; not verbatim OSI-approved MIT text)

  • Modified MIT: OpenAI disclaims ownership of generated content
  • WebText training data never released

Why it matters

GPT-2 set the template every later release followed: weights yes, data no — which is exactly why the rubric scores Access and Transparency separately.

Tier3HiDream-I1 (Full / Dev / Fast)HiDream.ai ImageMIT-licensed image weights with a published technical report behind them.APDTRGMIT License (code and model weights)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT License (code and model weights)

  • Apply the standard MIT notice and copyright notice to redistributed weights and code

Why it matters

A useful counterweight to the FLUX and Stability families: the same distributable-weight practice with a genuinely permissive license attached.

Tier3intfloat E5 (multilingual-e5, e5-mistral-7b-instruct)intfloat / Microsoft research EmbeddingsMIT-licensed embedding baseline still widely used for retrieval.APDTRGMIT (multilingual-e5 series, e5-mistral-7b-instruct, mmE5-mllama-11b-instruct)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT (multilingual-e5 series, e5-mistral-7b-instruct, mmE5-mllama-11b-instruct)

Why it matters

A second permissive embedding family, allowing the book to treat embedding openness as a solved licence problem with unsolved data problems.

Tier3Kimi-Audio (7B Instruct)Moonshot AI Speech & audioPermissively licensed audio foundation weights from Moonshot.APDTRGMIT for the weights; Apache-2.0 and MIT for the codeArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT for the weights; Apache-2.0 and MIT for the code

  • Standard MIT and Apache-2.0 obligations only

Why it matters

Noteworthy because the surrounding Kimi text families carry custom terms: the audio branch shows the same lab choosing ordinary permissive licensing.

Tier3Kokoro (82M)hexgrad Speech & audioAn 82-million-parameter Apache-2.0 voice model that runs almost anywhere.APDTRGApache-2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache-2.0

  • Standard Apache-2.0 obligations only

Why it matters

Shows that permissive weights plus small size can matter more than transparency for real-world openness: this one actually ships into products.

Tier3Microsoft Phi-4 family (Phi-4, mini, multimodal, reasoning, reasoning-plus, reasoning-vision-15B)Microsoft Text LLMsPractical small and mid-size models with unusually good artifact disclosure for a commercial lab.APDTRGMIT across the Phi-4 familyArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT across the Phi-4 family

  • Model cards state that nothing in the card is a restriction or modification of the license, but the release is version-specific in practice

Why it matters

Phi shows that smaller models can be more practically portable than frontier ones, and that a commercial lab can publish fine-tuning code and benchmark logs without opening the corpus.

Tier3Mistral Small 4 (119B, 2603)Mistral AI Text LLMsA specific Apache checkpoint in a portfolio with multiple license regimes.APDTRGApache License 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Mistral-Small-4-119B-2603 only, under its own Apache 2.0 license. Other Mistral releases are excluded.
Artifact kind
Trained model
Assessment recorded
Selected evidence checked

Apache License 2.0

  • Apache notice and license-preservation requirements apply. Other Mistral releases have separate terms and separate profiles.

Why it matters

Small 4's profile does not approve Mistral Large 4 or other checkpoints under different terms.

Tier3Mochi 1Genmo VideoAn Apache-2.0 video model whose practical limits are technical, not legal.APDTRGApache-2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache-2.0

  • Standard Apache-2.0 obligations only

Why it matters

Useful because the reasons to use it are hardware and resolution rather than licence terms — openness is not the binding constraint here.

Tier3Moonshot Kimi Linear (Kimi-Linear-48B-A3B base and instruct, KDA architecture)Moonshot AI Text LLMsKimi Delta Attention model released under plain MIT, unlike the modified MIT of Kimi K2.APDTRGMIT (HF license:mit tag on Kimi-Linear-48B-A3B-Base and Instruct)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT (HF license:mit tag on Kimi-Linear-48B-A3B-Base and Instruct)

Why it matters

A useful within-family licence contrast for the same vendor, showing that branding the whole vendor open-weight would have been wrong.

Tier3Moshi / Moshiko (with the Mimi codec)Kyutai Speech & audioFull-duplex speech dialogue with one of the most detailed training descriptions in the audio batch.APDTRGCC-BY 4.0 for model weights; MIT for the Python code and Apache-2.0 for the Rust backend — composite terms with no single OSI-approved software licence over the releaseArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
2/3 Yes, with caveats
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=2 D=3 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

CC-BY 4.0 for model weights; MIT for the Python code and Apache-2.0 for the Rust backend — composite terms with no single OSI-approved software licence over the release

  • Weight licenses are content licences rather than OSI-approved software licences, and some Kyutai checkpoints carry non-commercial variants
  • The model card states the system is intended for research use and advises against professional or advice-giving deployment
  • Attribution is required for redistributed weights

Why it matters

Shows that the transparency axis can score well even when the weight licence is a content licence — openness here is a spectrum across A, P, D, T, and R rather than one switch.

Tier3Nomic Embed (nomic-embed-text v1.5, nomic-embed-text-v2-moe)Nomic AI EmbeddingsApache-2.0 embeddings with Matryoshka truncation support.APDTRGApache 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0

Why it matters

Same permissive pattern, and a useful example of flexibility features (variable dimensions) shipped as part of the open artefact.

Tier3Open Jev (community reconstruction)Kye Gomez (community) Decision modelsThe community's answer to a closed decision model: the idea, rebuilt, minus the intelligence.APDTRGApache 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Method implementation
Assessment recorded

Apache 2.0

  • Random weights — explicitly not the production Jev model
  • Makes no claim of matching TypeSafe's published results

Why it matters

The paired control to Jev: when a category-defining model stays closed, the open ecosystem reconstructs the interface — and the reconstruction is honest that it is not equivalent. Interface compatibility is not model equivalence.

Tier3OpenAI gpt-oss (120b / 20b)OpenAI Text LLMsHighly usable open-weight model with Apache 2.0 weights but incomplete training openness.APDTRGApache License 2.0 on gpt-oss-120b and gpt-oss-20bArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded
Selected evidence checked

Apache License 2.0 on gpt-oss-120b and gpt-oss-20b

  • The accompanying usage-policy file expresses a safety aspiration and requires compliance with applicable law; the Apache license does not incorporate additional field-of-use conditions.

Why it matters

The release distinguishes Apache-licensed weights from OpenAI's hosted services, while leaving gaps in the training material available for independent reconstruction.

Tier3OpenCLIPML Foundations (open source community) VisionThe community's answer to CLIP's missing license.APDTRGMIT (code); released weights inherit upstream dataset termsArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
2/3 Yes, with caveats
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
3/3 Genuine community governance

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=2 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Training toolkit
Assessment recorded

MIT (code); released weights inherit upstream dataset terms

  • Weights trained on LAION and similar datasets carry non-commercial research terms

Why it matters

Shows open source re-implementations fixing what the original release left closed — at the cost of inheriting dataset-licensing problems.

Tier3pi0, pi0-FAST and pi0.5 (openpi)Physical Intelligence RoboticsOpen code and weights for a robot policy, with Gemma-licensed weights and closed pretraining data.APDTRGRepository code Apache 2.0; checkpoints are PaliGemma and Gemma derivatives and carry Gemma terms rather than plain Apache 2.0Artifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
2/3 Yes, with caveats
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=2 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Repository code Apache 2.0; checkpoints are PaliGemma and Gemma derivatives and carry Gemma terms rather than plain Apache 2.0

  • Weight terms are set by the Gemma licence, not the repo licence
  • The release is presented by the publisher as an experiment

Why it matters

Open robotics foundation models are real; the binding constraints are data provenance and backbone licensing, not access.

Tier3Qwen3-8BAlibaba Qwen team Text LLMsA specific Apache checkpoint that should not inherit the Max release's custom terms.APDTRGApache License 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Qwen3-8B only, under its own Apache 2.0 license; excludes Qwen3.8-Max and older custom-license checkpoints.
Artifact kind
Trained model
Assessment recorded
Selected evidence checked

Apache License 2.0

  • Ordinary Apache notice and license-preservation requirements.

Why it matters

Assessing a named checkpoint separates practical local control from training reconstruction and avoids family-wide license assumptions.

Tier3Qwen3-ASR (0.6B / 1.7B)Alibaba Qwen team Speech & audioApache-2.0 speech recognition with a 0.6B model small enough for on-device use.APDTRGApache-2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache-2.0

  • Standard Apache-2.0 obligations only

Why it matters

Shows speech recognition quietly becoming a permissively licensed commodity, which changes the strategic position of commercial transcription APIs.

Tier3Stable Audio (Open 1.0 / Open Small / 3.0)Stability AI Speech & audioOpen audio weights under the same community license that governs Stability's image releases.APDTRGStability AI Community LicenseArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
2/3 Yes, with caveats
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=2 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Stability AI Community License

  • Organizations with more than $1M in annual revenue must contact Stability AI for enterprise licensing
  • Published releases carry a license tag of 'other' rather than a named OSI licence, and the repositories are gated by agreement
  • Stability AI has repeatedly revised the audio release terms

Why it matters

Extends the Stability AI pattern from images into audio, showing that one community-license design can cover a whole product line across modalities.

Tier3V-JEPA 2 / V-JEPA 2.1Meta AI (FAIR) World modelsPermissively licensed video world model — the JEPA line's most open release.APDTRGMIT and Apache-2.0 across released checkpoints (GitHub code split MIT/Apache-2.0)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT and Apache-2.0 across released checkpoints (GitHub code split MIT/Apache-2.0)

  • Sibling jepa-wms research repo remains CC-BY-NC 4.0
  • V-JEPA 2.1 checkpoints distributed via download links, not yet the Meta HF org

Why it matters

Contradicts the assumption that Meta's JEPA family is uniformly non-commercial: the world-model generation moved to MIT/Apache while I-JEPA and the research repos stayed CC-BY-NC.

Tier3Wan 2.2 (T2V-A14B and companions)Alibaba Wan team VideoThe video model whose license file says Apache-2.0 and stops there.APDTRGApache-2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache-2.0

  • Standard Apache-2.0 obligations only; no revenue threshold, no territory carve-out, no field-of-use restriction

Why it matters

The useful control case for video: identical openness problems to its competitors, but no license caveat attached, which isolates the license variable from the transparency variable.

Tier3WhisperOpenAI Speech & audioThe most widely deployed open speech recognition stack, under a permissive license.APDTRGMIT License (code and weights; model cards also carry Apache-2.0 metadata)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT License (code and weights; model cards also carry Apache-2.0 metadata)

  • Standard MIT obligations only; no revenue threshold or use restriction

Why it matters

Demonstrates that permissive licensing plus a documented method can carry a model to universal deployment even without releasing the training data.

Tier3Writer Palmyra open line (Palmyra Small/Base, Palmyra-mini 1.5B family)Writer (Writer.com) Text LLMsAn enterprise AI company shipping genuinely permissive small models beside a proprietary platform.APDTRGApache 2.0 (palmyra-base and the palmyra-mini family; verify per release)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0 (palmyra-base and the palmyra-mini family; verify per release)

  • Verify each checkpoint; the enterprise Palmyra X-series and domain models (Med, Fin) are API/proprietary

Why it matters

The sponsored-open-source pattern in miniature: a commercial vendor gives away small permissive artifacts that funnel toward a closed product line — and unlike Gemma, the licenses are clean.

Tier3Z-Image (base / Turbo)Alibaba Tongyi Lab ImageApache-2.0 image weights that trade capability for speed without adding license conditions.APDTRGApache-2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Independent access and deployment pass; T<2 or R<2 prevents Tier 4. A=3 P=3 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache-2.0

  • Standard Apache-2.0 obligations only: retain notices and state changes

Why it matters

Demonstrates that an OSI-approved weight license can coexist with essentially no training transparency — permissiveness and reproducibility are separate axes.

Tier3Z.AI GLM-5.3 FlashZ.AI (Zhipu AI) Text LLMsMIT permissions on a named Flash checkpoint, distinct from its custom-licensed sibling.APDTRGMIT LicenseArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

A=3 P=3 D=3 T=1 R=1. Training reconstruction evidence is below the Tier 4 threshold; governance remains separate.

Scope and terms

Assessment scope
GLM-5.3-Flash only, under its own MIT license; excludes GLM-5.3 and its custom security-review clause.
Artifact kind
Trained model
Assessment recorded
Selected evidence checked

MIT License

  • Ordinary MIT copyright and permission notice preservation.

Why it matters

Separate profiles prevent a larger sibling's conditions from being attributed to an MIT checkpoint.

Tier2

Restricted or deployment-limited

A=3 and D>=2, but Permission is below 2 or Deployment is below 3.

18 profiles
Tier2AlphaFold 3Google DeepMind and Isomorphic Labs ScienceOpen code, restricted weights: the sharpest licence contrast in the batch.APDTRGCode Apache 2.0; parameters under AlphaFold 3 Model Parameters Terms of Use plus the Model Parameters Prohibited Use PolicyArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Code Apache 2.0; parameters under AlphaFold 3 Model Parameters Terms of Use plus the Model Parameters Prohibited Use Policy

  • Non-commercial use only, including research on behalf of commercial organizations
  • Weights may not be published or shared outside the requesting organization
  • Output may not be used to train similar structure-prediction models
  • Revocable licence; access can be terminated on breach

Why it matters

Same lab, adjacent year, radically different permission profile, which is exactly what the word open has to survive.

Tier2AudioCraft / MusicGen (small, medium, large, melody)Meta Speech & audioThe reference open music-generation stack, whose weights are non-commercial.APDTRGMIT for the AudioCraft code; CC-BY-NC 4.0 for the MusicGen model weightsArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT for the AudioCraft code; CC-BY-NC 4.0 for the MusicGen model weights

  • The published MusicGen weights are non-commercial: CC-BY-NC 4.0 prohibits commercial use of the licensed material
  • Commercial deployment of MusicGen requires separate permission from Meta
  • Code and weights carry different licences in the same repository

Why it matters

The clearest case in the dataset of code openness and weight openness diverging inside a single release — you may run and modify it, but you may not sell what it makes.

Tier2BigScience BLOOMBigScience Workshop Text LLMsOpen weights wrapped in a purpose-built responsible-AI licence.APDTRGBigScience RAIL License v1.0Artifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
3/3 Genuine community governance

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

BigScience RAIL License v1.0

  • custom Responsible AI License with a prohibited-use list (discrimination, medical advice, law-enforcement and justice-administration uses barred)
  • permissive grant modified from Apache 2.0

Why it matters

The origin point of the RAIL licence stack that later open-weight releases copied, including the pattern of attaching use restrictions to otherwise permissive terms.

Tier2Cohere Command A / Command R / Aya open weightsCohere (Cohere Labs) Text LLMsResearch-grade open weights that are explicitly non-commercial.APDTRGCC-BY-NC 4.0 with Cohere Labs Acceptable Use Addendum (c4ai-cc-by-nc-license on Command R; CC-BY-NC-4.0 tag on Command A and Aya)Artifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

CC-BY-NC 4.0 with Cohere Labs Acceptable Use Addendum (c4ai-cc-by-nc-license on Command R; CC-BY-NC-4.0 tag on Command A and Aya)

  • non-commercial only
  • Cohere Labs Acceptable Use Policy incorporated by reference
  • gated repositories requiring accepting the license agreement
  • use-based prohibited-use list

Why it matters

The clearest corporate example of research-only openness: usable, deployable, and legally closed to commercial use.

Tier2DeepSeek-Coder and DeepSeek-Coder-V2DeepSeek CodeSame lab, different licence: the code line ships under DeepSeek custom terms rather than the MIT used for V3 and R1.APDTRGDeepSeek License Agreement v1.0 (model licence), distinct from the MIT used by DeepSeek V3/R1Artifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
2/3 Yes, with caveats
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=2 R=2. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

DeepSeek License Agreement v1.0 (model licence), distinct from the MIT used by DeepSeek V3/R1

  • custom permissive-with-restrictions licence: redistribution allowed with the same use-based restrictions carried into derivatives, plus prohibited uses (illegal content, PII, defamation, discrimination, military use, fully automated decisions affecting legal rights)

Why it matters

Materially distinct from the DeepSeek text-family entry in the dataset; shows that openness must be read per model, not per vendor.

Tier2FLUX 3 Image Klein (5.6B)Black Forest Labs ImageTop-of-chart open image weights with a revenue threshold written into the license.APDTRGFLUX Kommunity License v1.0Artifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

FLUX Kommunity License v1.0

  • Non-commercial use only, except that Qualifying Users may generate commercial output if they have less than US$5M in gross annualized revenue
  • Acceptable Use Policy and trademark conditions apply
  • The Qwen3-VL-4B-Instruct text encoder inside the package is separately Apache-2.0

Why it matters

The clearest current example of 'open weights' plus a hard commercial ceiling: the license names a dollar figure, so the openness question becomes a business-model question.

Tier2HunyuanVideo (1.5 and earlier)Tencent Hunyuan VideoOpen video weights whose license does not apply in three of the largest markets.APDTRGTencent Hunyuan Community LicenseArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
2/3 Yes, with caveats
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=2 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Tencent Hunyuan Community License

  • The license does not apply in the European Union, the United Kingdom, or South Korea, and is expressly limited to a Territory that excludes them
  • Additional commercial terms kick in for licensees with more than 100 million monthly active users
  • A separate Acceptable Use Policy is incorporated as Exhibit A

Why it matters

Territory clauses are the least discussed form of restriction: a licence can be free and open everywhere except where the publisher decides it should not be.

Tier2I-JEPAMeta AI (FAIR) VisionThe image JEPA pioneer, under a non-commercial license.APDTRGCC-BY-NC 4.0 (code and weights)Artifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

CC-BY-NC 4.0 (code and weights)

  • Non-commercial use only

Why it matters

Shows the original JEPA release was open-access, not open source — the license moved before the architecture became famous.

Tier2Meta Llama 4 Scout and MaverickMeta Text LLMsDownloadable multimodal models with release-specific license and deployment conditions.APDTRGLlama 4 Community License and incorporated Llama 4 Acceptable Use PolicyArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
2/3 Yes, with caveats
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=2 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Llama 4 Scout and Maverick under the April 5, 2025 Community License and incorporated AUP. Muse and other Llama generations are excluded.
Artifact kind
Trained model
Assessment recorded

Llama 4 Community License and incorporated Llama 4 Acceptable Use Policy

  • Products above 700M monthly active users must request a separate Meta license, granted at Meta's sole discretion
  • Mandatory 'Built with Llama' attribution and a 'Llama' naming prefix for derivatives
  • Multimodal rights are withheld from individuals and companies based in the European Union
  • Llama 4 repositories are gated on Hugging Face behind license acceptance

Why it matters

Llama 4 shows why eligible access, technical deployment and legal permission must be assessed separately.

Tier2Mistral Large 4Mistral AI Text LLMsEuropean frontier-scale open weights with a commercial carve-out.APDTRGMistral Ecosystem Open-Weight License, a modified MITArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Mistral Ecosystem Open-Weight License, a modified MIT

  • Businesses that operate a Model-as-a-Service offering must obtain a separate commercial license from Mistral before using the model commercially
  • Attribution notice and license text must be retained in copies and substantial portions

Why it matters

Large 4 marks Mistral's flagship moving off Apache 2.0, and its MaaS carve-out joins the 2026 pattern of vendor-set revenue and business-model conditions attached to nominally open weights.

Tier2Mistral Medium 3.5Mistral AI Text LLMsOpen weights whose license switches off for larger companies.APDTRGModified MIT with a company-wide monthly revenue thresholdArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Modified MIT with a company-wide monthly revenue threshold

  • If the licensee's consolidated global monthly revenue exceeded $20M in the preceding month, no rights under the license may be exercised and a commercial license must be sought
  • The threshold applies to total company revenue, not only to the service built on the model

Why it matters

Medium 3.5 is the sharpest test of the 'downloadable means unrestricted' assumption: the weights are open and the license is a modified MIT, yet the grant evaporates for any company above a revenue line.

Tier2Nex AGI Nex (Nex-N2.6; the N1 and N2.x line)Nex AGI (an alliance initiated by the Shanghai Innovation Institute) Text LLMsApache-2.0 post-training of another lab's open base, where the openness stops at the checkpoint.APDTRGApache License 2.0 (HF license:apache-2.0 tag on Nex-N2.6; consistent across the N1 and N2.x releases)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
2/3 Yes, with caveats
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=3 D=2 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache License 2.0 (HF license:apache-2.0 tag on Nex-N2.6; consistent across the N1 and N2.x releases)

Why it matters

Nex AGI markets an open ecosystem while publishing only weights: no data, no code, no recipe, no governance. It shows how an open license on a borrowed base can carry an openness reputation the artifacts do not.

Tier2OpenAI CLIPOpenAI VisionMIT code around weights nobody ever licensed.APDTRGMIT (code); model card carries no license tag for the weightsArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

MIT (code); model card carries no license tag for the weights

  • Model card: 'out of scope' for all deployed use cases; no explicit weight license grant
  • Training corpus (WebImageText, 400M pairs) never released

Why it matters

The historical precedent for the whole weights-without-license pattern that later open-weight releases formalized.

Tier2Prism ML Bonsai (1-bit and ternary Bonsai, Bonsai 2 27B)Prism ML Text LLMsEfficiency research released under Apache 2.0 but gated on unreleased kernels and an opaque base.APDTRGApache 2.0 (HF license:apache-2.0 on Bonsai-8B-gguf, Bonsai-8B-mlx-1bit, Bonsai-8B-unpacked and Ternary-Bonsai releases)Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
2/3 Yes, with caveats
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=3 D=2 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0 (HF license:apache-2.0 on Bonsai-8B-gguf, Bonsai-8B-mlx-1bit, Bonsai-8B-unpacked and Ternary-Bonsai releases)

Why it matters

Shows that openness of the artefact and openness of the training story can fail independently in the same release.

Tier2Qwen3.8-Max (2.4T-A95B)Alibaba Qwen team Text LLMsDownloadable weights with commercial thresholds that do not apply to Qwen3-8B.APDTRGQwen3.8-Max LicenseArtifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Qwen3.8-2.4T-A95B under the Qwen3.8-Max License only.
Artifact kind
Trained model
Assessment recorded
Selected evidence checked

Qwen3.8-Max License

  • Model-name display above the stated MAU or monthly-revenue thresholds.
  • Qualifying Model-as-a-Service or AI Work Assistant businesses above $50M aggregate revenue over twelve consecutive months require a separate license.

Why it matters

The artifact can be independently hosted, but legal permission depends on the exact license and the licensee's business.

Tier2Thinking Machines Inkling (Inkling 2 Small, 594B MoE; Inkling 975B and Inkling-Small from July 2026)Thinking Machines Lab Text LLMsThe first Apache-2.0 frontier-class weights release from a US lab, and it arrives on the lab's own safety schedule.APDTRGApache License 2.0 on the downloadable weights (NVFP4 and MXFP8 checkpoints); a separate Acceptable Use Policy governs only the hosted Tinker APIArtifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
2/3 Yes, with caveats
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
2/3 Real but bounded participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=3 D=2 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache License 2.0 on the downloadable weights (NVFP4 and MXFP8 checkpoints); a separate Acceptable Use Policy governs only the hosted Tinker API

Why it matters

The staged release — closed on Tinker, then downloadable after external red-teaming — is now the template for how a well-funded Western lab ships openness: real weights, real permissiveness, on a timeline the company controls.

Tier2Z.AI GLM-5.3Z.AI (Zhipu AI) Text LLMsOpen-weight frontier coding model, not a full open-source AI release under the strict rubric.APDTRGGLM-5.3 License (custom, with a qualifying Model-as-a-Service security-review requirement)Artifact license: not OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
1/3 Limited / partial
DDeployment
3/3 Strong / exemplary
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=1 D=3 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
GLM-5.3 only. GLM-5.3 Flash and earlier MIT releases are outside this assessment.
Artifact kind
Trained model
Assessment recorded

GLM-5.3 License (custom, with a qualifying Model-as-a-Service security-review requirement)

  • Model-as-a-Service operators with more than $10B aggregate 12-month revenue must pass a Z.AI security review whose scope and method Z.AI determines
  • Release-specific terms; the license must be read rather than inferred from open branding

Why it matters

Z.AI explicitly uses the open-weight label and discusses vulnerability-discovery training, while the release provides vendor-controlled security review and no OLMo-level reproducibility - the label and the evidence point in different directions.

Tier2Zyphra ZAYA1 (ZAYA1-8B, ZAYA1-VL-8B, ZAYA1 base)Zyphra Text LLMsPermissive licence undercut by a deployment dependency on unreleased upstream patches.APDTRGApache 2.0Artifact license: OSI-approved

Six dimensions

AAccess
3/3 Strong / exemplary
PPermission
3/3 Strong / exemplary
DDeployment
2/3 Yes, with caveats
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Permission/deployment gate: P<2 or D<3. A=3 P=3 D=2 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Apache 2.0

Why it matters

Shows that a standard licence does not guarantee open deployment when the runtime lives in private forks.

Tier1

Access-limited

Some runnable artifacts are public, but Access is below 3 or Deployment is below 2.

1 profile
Tier1DINOv3Meta AI (FAIR) VisionGated weights under a bespoke agreement — open in marketing, licensed in substance.APDTRGCustom DINOv3 LicenseArtifact license: not OSI-approved

Six dimensions

AAccess
2/3 Yes, with caveats
PPermission
2/3 Yes, with caveats
DDeployment
3/3 Strong / exemplary
TTransparency
2/3 Yes, with caveats
RReproducibility
1/3 Limited / partial
GGovernance
1/3 Cosmetic participation

Tier rationale

Access/deployment gate: A<3 or D<2. A=2 P=2 D=3 T=2 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Trained model
Assessment recorded

Custom DINOv3 License

  • Gated HF access with manual approval and contact-info sharing
  • Royalty-free but non-transferable grant; redistribution only under the same agreement
  • Mandatory 'Built with DINOv3' attribution
  • Trade-control and military-use exclusions; Meta may amend terms unilaterally
  • No commercial-use clause found in the license text

Why it matters

The clearest vision-side example of openness retreating between generations: DINOv2 Apache, DINOv3 gated custom terms.

Tier0

Proprietary service

Access and Deployment both score 0: no independently runnable release is available.

4 profiles
Tier0ESM3 flagship (98B and 7B)Chan Zuckerberg Biohub, formerly EvolutionaryScale ScienceFrontier protein model, API only.APDTRGBiohub platform terms and acceptable use policy; no published licence for flagship weightsArtifact license: not OSI-approved

Six dimensions

AAccess
0/3 No
PPermission
0/3 No
DDeployment
0/3 No
TTransparency
1/3 Limited / partial
RReproducibility
0/3 No
GGovernance
0/3 No participation path

Tier rationale

No independently runnable artifact: A=0 and D=0. A=0 P=0 D=0 T=1 R=0. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Hosted service
Assessment recorded

Biohub platform terms and acceptable use policy; no published licence for flagship weights

  • API and platform access only
  • Acceptable Use Policy at biohub.org

Why it matters

Openness here is graded rather than binary: an open small model and a closed flagship under one roof.

Tier0Google DeepMind Genie 3 / Genie 4Google DeepMind World modelsThe frontier world model, fully closed.APDTRGNone published — closed research previewArtifact license: not OSI-approved

Six dimensions

AAccess
0/3 No
PPermission
0/3 No
DDeployment
0/3 No
TTransparency
1/3 Limited / partial
RReproducibility
0/3 No
GGovernance
0/3 No participation path

Tier rationale

No independently runnable artifact: A=0 and D=0. A=0 P=0 D=0 T=1 R=0. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Hosted service
Assessment recorded

None published — closed research preview

  • No weights, no code, no license
  • Genie 4 (Oct 2026) rolling out to trusted testers, then AI Ultra / Project Genie, then Gemini API / AI Studio

Why it matters

The tier-0 control for the world-model category: the most capable system in the space ships zero artifacts.

Tier0OpenAI Five (Dota 2)OpenAI Decision & gamingThe famous game-playing agent that was never released.APDTRGNo released code or weights; paper and blog post onlyArtifact license: not OSI-approved

Six dimensions

AAccess
0/3 No
PPermission
0/3 No
DDeployment
0/3 No
TTransparency
1/3 Limited / partial
RReproducibility
1/3 Limited / partial
GGovernance
0/3 No participation path

Tier rationale

No independently runnable artifact: A=0 and D=0. A=0 P=0 D=0 T=1 R=1. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Hosted service
Assessment recorded

No released code or weights; paper and blog post only

  • Nothing published to license or reuse; it ran as an internal research artifact

Why it matters

The negative control, a landmark result whose openness stops at the paper, which is where open usually stops.

Tier0TypeSafe Jev (System One models)TypeSafe AI Decision modelsThe open interface around a closed intelligence.APDTRGNone published for the model — hosted service (early access)Artifact license: not OSI-approved

Six dimensions

AAccess
0/3 No
PPermission
0/3 No
DDeployment
0/3 No
TTransparency
1/3 Limited / partial
RReproducibility
0/3 No
GGovernance
1/3 Cosmetic participation

Tier rationale

No independently runnable artifact: A=0 and D=0. A=0 P=0 D=0 T=1 R=0. G is displayed separately.

Scope and terms

Assessment scope
Only the named artifacts and release versions listed in this profile; no score applies to other releases from this publisher.
Artifact kind
Hosted service
Assessment recorded

None published for the model — hosted service (early access)

  • No weights, no architecture publication, no technical paper, no license
  • Hosted only via console/API (also listed on OpenRouter as typesafe/jev-1.13 and jev-1.14)
  • Python and JavaScript SDKs and the workflow-eval harness ARE open source — the interface is open, the model is not

Why it matters

The sharpest new form of the book's question: a decision model whose SDKs, eval harness, and type guarantees are all inspectable while the model itself is a black-box service. 'Open source' is doing zero work here — the openness lives entirely at the boundary.

Varies

Family overviews: terms vary

Reference entries outside the tier spectrum. Follow a named release before making a permission or deployment decision.

17 overviews
VariesBaidu ERNIE (ERNIE 4.5 open-weight line, ERNIE 5.5 Flash Base)Baidu Text LLMsOpen-weight ERNIE exists under Apache 2.0, but the newest release arrives under a bespoke licence.Terms varyMixed - Apache 2.0 for ERNIE-4.5-21B-A3B-Thinking and the ERNIE-4.5-VL checkpoints; custom ERNIE 5.5 Flash Base Model License Agreement v1.0 (dated 5 November 2026) for ERNIE-5.5-Flash-BaseTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Mixed - Apache 2.0 for ERNIE-4.5-21B-A3B-Thinking and the ERNIE-4.5-VL checkpoints; custom ERNIE 5.5 Flash Base Model License Agreement v1.0 (dated 5 November 2026) for ERNIE-5.5-Flash-Base

  • the 5.5 licence is a custom use-restricted agreement with a prohibited-use list including military use, high-stakes automated decisions and professional practice without authorisation
  • definition of licensed materials explicitly excludes unreleased training code and data
  • the 5.5 licence excludes the EU, UK and South Korea from the licensed territory outright
  • the 5.5 licence prohibits using the model or its outputs to improve any other AI model, and requires 'Powered by ERNIE' attribution with revenue-tiered notice to Baidu

Why it matters

Shows a large vendor moving from proprietary API-only to a split licence portfolio in a single year.

VariesCogVideoX (2b, 5b, 1.5, I2V variants)Zhipu AI / Z.ai (THUDM, zai-org) VideoOpen video weights whose commercial use is a registration away from a monthly cap.Terms varyThe CogVideoX License (with Apache-2.0 on the 2b checkpoint only)Terms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

The CogVideoX License (with Apache-2.0 on the 2b checkpoint only)

  • Commercial use requires registering and obtaining a basic commercial license
  • Commercial use is capped at 1 million visits per month before a further commercial license must be negotiated
  • Military, illegal, and national-security-related uses are prohibited, and the license notes it may be replaced by a more comprehensive version
  • Governed by PRC law with Beijing Haidian District jurisdiction

Why it matters

Registration-plus-cap is a distinct licence design: nothing is hidden, but the right to serve commercially is metered.

VariesDepth Anything V2Depth Anything project (HKU, TikTok) VisionOpen except for the checkpoints you actually want.Terms varyMixed: Small = Apache 2.0; Base/Large/Giant = CC-BY-NC 4.0Terms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Mixed: Small = Apache 2.0; Base/Large/Giant = CC-BY-NC 4.0

  • Non-commercial on all but the Small variant

Why it matters

A permissiveness gradient inside a single release: the smallest model is free, the capable ones are not.

VariesFalcon family (7B/40B, Falcon 3, Falcon-H1 / H1R)Technology Innovation Institute (TII), Abu Dhabi Text LLMsHistorically important model with meaningful but release-variable transparency.Terms varyApache 2.0 for Falcon 7B/40B; TII Falcon LLM License 2.0 for Falcon 3 and the Falcon-H1 familyTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Apache 2.0 for Falcon 7B/40B; TII Falcon LLM License 2.0 for Falcon 3 and the Falcon-H1 family

  • TII Falcon LLM License 2.0 is Apache-2.0-derived but adds an Acceptable Use Policy, a mandatory attribution notice, and commercial obligations
  • Hosted use of Falcon-180B may require a separate license
  • License varies by version and the OSI has disputed the open-source claim

Why it matters

Falcon is an early sovereign-AI comparison point and a clean case for why 'open source' has to be checked at the exact checkpoint, since one family spans plain Apache 2.0 and a custom license with conditions.

VariesFLUX.1 family (schnell / dev / pro)Black Forest Labs ImageThe open-weight image family whose own release table shows three different licenses in one row.Terms varyMixed by variant: FLUX.1 [schnell] Apache-2.0; FLUX.1 [dev] FLUX.1-dev Non-Commercial License; FLUX.1 [pro] proprietary API onlyTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Mixed by variant: FLUX.1 [schnell] Apache-2.0; FLUX.1 [dev] FLUX.1-dev Non-Commercial License; FLUX.1 [pro] proprietary API only

  • The official model table lists different licenses per variant, so the family label hides the terms of the variant you actually download
  • FLUX.1 [dev] weights are released under a non-commercial license
  • FLUX.1 [pro] has no downloadable weights at all

Why it matters

A single name can carry Apache-2.0, a non-commercial custom license, and no weights at all — so openness has to be checked per variant, not per family.

VariesFLUX.2 [klein] (4B / 9B)Black Forest Labs ImageTwo variants with one genuinely permissive license and one non-commercial license.Terms varyApache-2.0 for FLUX.2 [klein] 4B; FLUX Non-Commercial License for FLUX.2 [klein] 9BTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Apache-2.0 for FLUX.2 [klein] 4B; FLUX Non-Commercial License for FLUX.2 [klein] 9B

  • The 9B weights are gated behind acceptance of the FLUX Non-Commercial License and the BFL Acceptable Use Policy
  • The smaller 4B variant is genuinely Apache-2.0, so terms differ within the family

Why it matters

A direct test of whether a family name means anything: the same 'open weights' label covers Apache-2.0 weights and non-commercial weights in the same release.

VariesJina AI embeddings (jina-embeddings-v3, jina-embeddings-v4)Jina AI EmbeddingsTechnically open embedding weights under research-only licences, including a corrected licence mistake.Terms varyCC-BY-NC 4.0 for v3; Qwen Research License for v4 (research-only, non-commercial; the model card states v4 was first published under CC-BY-NC 4.0 by mistake and corrected to the Qwen Research License because it derives from Qwen2.5-VL-3B)Terms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

CC-BY-NC 4.0 for v3; Qwen Research License for v4 (research-only, non-commercial; the model card states v4 was first published under CC-BY-NC 4.0 by mistake and corrected to the Qwen Research License because it derives from Qwen2.5-VL-3B)

  • non-commercial use only
  • commercial use requires a separate licence request to Alibaba Cloud under the Qwen Research License
  • attribution and Notice-file obligations on redistribution

Why it matters

Shows a licence error being corrected in public, and the practical consequence of inheriting a base model licence.

VariesLTX-Video / LTX-2 and laterLightricks VideoOpen video weights with a revenue threshold and a prohibited-use annex.Terms varyLTXV Open Weights License 0.X; LTX-2.x Community License for later releasesTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

LTXV Open Weights License 0.X; LTX-2.x Community License for later releases

  • Entities with at least $10,000,000 in annual revenue are Commercial Entities that must obtain a separate paid Commercial Use Agreement
  • Attachment A lists prohibited uses, and unauthorized commercial use by a Commercial Entity carries liquidated damages
  • The training Data is explicitly not licensed under the agreement
  • Transfer of fine-tunes may require a paid license under the LTX-2.x Community License

Why it matters

The annex structure is the point: the license grants broad freedoms and then subtracts a long list of uses, which is how 'open weights' releases quietly reintroduce acceptable-use policy.

VariesMiniMax M2 / M2.7 / M3MiniMax (MiniMax AI) Text LLMsFrontier-scale MoE family whose openness tightens release over release, from modified MIT to outright non-commercial.Terms varyModified MIT (M2); non-commercial MIT-style (M2.7); MiniMax Community License (M3) - verify per releaseTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Modified MIT (M2); non-commercial MIT-style (M2.7); MiniMax Community License (M3) - verify per release

  • M2 adds a branding clause only for products over 100M MAU or 30M USD ARR
  • M2.7 permits non-commercial use only and requires prior written authorization for any commercial use
  • M3 grants non-commercial use and requires Built with MiniMax M3 display plus a notice or authorization above 20M USD yearly revenue

Why it matters

Shows that open weights can be shipped under a source-available reading of the word open, with the same vendor marking M2 open-source in a README while its own M2.7 LICENSE restricts commercial use.

VariesMoonshot Kimi (K2, K2.5, K2.6, K3, K3.1)Moonshot AI Text LLMsFrontier-scale open-weight family with custom or modified terms and incomplete training transparency.Terms varyModified MIT for the K2 family; custom Kimi K3 License for Kimi K3; custom Kimi K3.1 Pro License for K3.1 ProTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Modified MIT for the K2 family; custom Kimi K3 License for Kimi K3; custom Kimi K3.1 Pro License for K3.1 Pro

  • K3 requires a separate agreement with Moonshot for Model-as-a-Service businesses above $20M aggregate revenue over any 12 months
  • K3 requires the model name displayed prominently above 100M MAU or $20M monthly revenue
  • K2-family Modified MIT adds attribution and display conditions and a synthetic-data clause
  • K3.1 Pro (weights public early November 2026, 1M-token context, called Moonshot's most capable 3T-class open-weight model) ships under its own custom Kimi K3.1 Pro License rather than an OSI-approved license

Why it matters

Kimi combines frontier-scale capability and practical self-hosting with a license and transparency profile that is not the same as fully open source, and its 2026 scale makes the MaaS condition bite far more companies than it would for a small model.

VariesNVIDIA Cosmos 3NVIDIA World modelsInfrastructure-grade world model under a promising-but-unapproved open license.Terms varyNVIDIA Open Model License / OpenMDW-1.1Terms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

NVIDIA Open Model License / OpenMDW-1.1

  • OpenMDW-1.1 submitted to OSI license review 2026-08-14, not approved

Why it matters

The overview compares NVIDIA releases whose artifact licenses differ. Approval of a particular license does not establish that every model release meets OSAID requirements; inspect the exact checkpoint and accompanying resources.

VariesNVIDIA Isaac GR00T (N1 and N1.5)NVIDIA RoboticsOpen robot foundation model whose first weight releases are non-commercial.Terms varyCode Apache 2.0; N1 and N1.5 weights under the NVIDIA one-way non-commercial licence per their model cards; newer N1.7 weights under the NVIDIA Open Model LicenseTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Code Apache 2.0; N1 and N1.5 weights under the NVIDIA one-way non-commercial licence per their model cards; newer N1.7 weights under the NVIDIA Open Model License

  • N1 and N1.5 weights are non-commercial
  • NVIDIA Open Model License requires an attribution notice and Cosmos branding for Cosmos-derived work

Why it matters

Robotics repeats the text-LLM pattern: open code and checkpoints, with permission lagging capability.

VariesNVIDIA Nemotron (3, 3.5)NVIDIA Text LLMsCommercial model family with unusually deep data and recipe disclosure.Terms varyNVIDIA Nemotron Open Model License for the Nemotron 3 line; OpenMDW-1.1 (Linux Foundation) for Nemotron 3.5 releasesTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

NVIDIA Nemotron Open Model License for the Nemotron 3 line; OpenMDW-1.1 (Linux Foundation) for Nemotron 3.5 releases

  • OpenMDW-1.1 has been submitted to the OSI but is not yet approved
  • Terms are published per asset, so each checkpoint must be checked

Why it matters

Nemotron shows that a commercial hardware company can choose to publish beyond weights, and its move to OpenMDW-1.1 shows how a new permissive license family is being adopted for AI model distributions.

VariesQwen family (Qwen3, Qwen3.5, Qwen3.8, Qwen3.8-Max)Alibaba Qwen team Text LLMsBroadly deployable open-weight family whose top tier now carries a custom license.Terms varyApache 2.0 across Qwen3, Qwen3.5, and Qwen3.8-27B; custom Qwen3.8-Max License for the 2.4T-A95B Max tier; older Qwen2.5 checkpoints retain a custom Qwen licenseTerms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Apache 2.0 across Qwen3, Qwen3.5, and Qwen3.8-27B; custom Qwen3.8-Max License for the 2.4T-A95B Max tier; older Qwen2.5 checkpoints retain a custom Qwen license

  • Qwen3.8-Max License requires displaying the model name above 100M MAU or $20M monthly revenue
  • Qwen3.8-Max License requires a separate Qwen license for Model-as-a-Service or AI Work Assistant businesses above $50M TTM aggregate revenue
  • Qwen2.5 checkpoints carry an MAU threshold and derived-model attribution conditions

Why it matters

Qwen shows how a model can be highly portable and permissively licensed at the small end while the frontier tier moves to revenue-conditioned terms, which is the clearest 2026 case of a family split across two license regimes.

VariesQwen-Image (base, Edit, 2.1, 3.0)Alibaba Qwen team ImageA strong Apache-2.0 image model whose successor reclassified itself as research-only and whose newest version has no weights.Terms varyApache-2.0 for Qwen-Image and Qwen-Image-Edit; Qwen Research License Agreement for Qwen-Image-2.1Terms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Apache-2.0 for Qwen-Image and Qwen-Image-Edit; Qwen Research License Agreement for Qwen-Image-2.1

  • The 2.1 release (September 2026) is governed by a Qwen Research License Agreement rather than Apache-2.0
  • Qwen-Image-3.0 ships no weights at all, so the newest capability is not downloadable
  • Permissions must be checked at each release, not at the family level

Why it matters

Shows how a permissive open-weight launch can be quietly replaced by a research license and an API-only roadmap — the label survives, the terms change.

VariesStable Diffusion (1.5 / SDXL / 3 / 3.5)Stability AI ImageThe most-deployed open image family, whose license tightened generation by generation.Terms varyCreativeML Open RAIL-M / Open RAIL++-M for 1.5 and SDXL; Stability AI Community License for 3 and 3.5Terms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

CreativeML Open RAIL-M / Open RAIL++-M for 1.5 and SDXL; Stability AI Community License for 3 and 3.5

  • Use-based restrictions remain in the RAIL licenses, so they are not OSI-approved open source
  • Under the Community License, organizations with more than $1M in annual revenue must negotiate an enterprise license
  • Terms changed across generations inside the same family

Why it matters

Stability AI's own community license is the reference case for how an incumbent converts an open-weight release into a licensed-gated one without ever closing the weights.

VariesUpstage Solar (SOLAR-10.7B base and instruct, Solar-Open and Solar-Open2)Upstage Text LLMsSame family, three different licences - Apache 2.0, CC-BY-NC and a branded Apache derivative.Terms varyMixed across releases - Apache 2.0 for SOLAR-10.7B-v1.0 base; CC-BY-NC 4.0 for SOLAR-10.7B-Instruct-v1.0; custom Upstage Solar Licence (Apache 2.0 derived with a branding condition) for Solar-Open and Solar-Open2Terms vary by release

Family overview

No aggregate scores or tier: release terms differ within this family.

Why no aggregate tier

No aggregate tier: the family includes releases with materially different permissions or access. Historical numerical assignments remain in dated snapshots.

Scope and terms

Assessment scope
Mixed-license family overview. Terms vary between the named releases; there is no family-wide permission score or tier. Follow the exact release license in the source list.
Artifact kind
Trained model
Assessment recorded

Mixed across releases - Apache 2.0 for SOLAR-10.7B-v1.0 base; CC-BY-NC 4.0 for SOLAR-10.7B-Instruct-v1.0; custom Upstage Solar Licence (Apache 2.0 derived with a branding condition) for Solar-Open and Solar-Open2

  • licence changes between checkpoints
  • the Upstage Solar Licence permits commercial use but requires derivative models to be named Solar and to display Built with Solar

Why it matters

Retained as an entry with the explicit caveat that Upstage openness must be checked at the exact checkpoint and licence level.

Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-15. Each profile records its scope and assessment date; a dataset snapshot does not mean every source was checked that day. Scores are qualitative and source-linked. The dataset and its history ↗ · Compare two releases →