← The openness index

text-llm · Openness profile

Mistral Medium 3.5

Mistral AI · snapshot 2026-11-07

Open weights with a modified MIT license that ties commercial use to company size.

Tier 2 Restricted open-weight

What the publisher claims: open-weight, open models

The one-liner: Open weights whose license switches off for larger companies.

The six questions

A AccessScore 3 of 3

Weights are downloadable from Hugging Face as Mistral-Medium-3.5-128B.

P PermissionScore 1 of 3

The modified MIT grant withdraws itself for any company whose consolidated global monthly revenue exceeded $20M in the preceding month, so a substantial class of users has no license at all and must renegotiate; that is a meaningful restriction rather than a caveat.

D DeploymentScore 3 of 3

Runs independently on chosen hardware with standard open inference stacks, once the license question is cleared.

T TransparencyScore 1 of 3

A model card and technical material are published; training data and the full recipe are not.

R ReproducibilityScore 1 of 3

Limited: no published corpus or reproducible training recipe.

G Governance & ParticipationScore 1 of 3

Medium 3.5 releases weights, a model card, and deployment guidance with no public training or evaluation repository attached, so outside developers can only raise issues with Mistral.

License

Modified MIT with a company-wide monthly revenue threshold — license source ↗

What is actually released

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.

Sources

Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-07 ↗. Scores are qualitative readings of the cited evidence, not verdicts; tier measures openness, not intelligence.

← The openness indexJoin the newsletter ↗