Tier 3 Practical open-weight
What the publisher claims: open models, open-weight, open-source
The one-liner: A major lab family that moved to a standard license without opening the training system.
The six questions
All five Gemma 4 sizes, in both pre-trained and instruction-tuned variants, are downloadable from Hugging Face and GitHub.
Gemma 4 weights are under plain Apache 2.0, an OSI-approved grant with no revenue threshold, use policy, or attribution condition attached to the weights themselves.
Designed for on-device and server deployment, from phone-class E2B models up to 31B and 26B-A4B checkpoints, with broad support across local runtimes.
The model card documents architecture, context windows, benchmark tables, and safety evaluations, and a technical report exists, but the training data and recipe are not published.
Limited to moderate: an equivalent model can be approximated from the paper but not rebuilt from published data.
Gemma 4's Apache-2.0 weights and model card are published with no development repository, roadmap, or community forum, so Google accepts no outside contribution to the model itself.
License
Apache 2.0 for the Gemma 4 weights, a change from the custom Gemma terms used for earlier generations OSI-approved — license source ↗
- 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
What is actually released
- weights (E2B, E4B, 12B Unified, 26B A4B MoE, 31B; pre-trained and instruction-tuned)
- model card with architecture and benchmark tables
- technical report (arXiv 2607.02770)
- safety evaluation results
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.
Sources
- https://ai.google.dev/gemma/docs/core/model_card_4
- https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4
- https://ibl.ai/blog/gemma-4-apache-2-open-weight-models-enterprise
- https://arxiv.org/abs/2607.02770
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.