Gemma 4: what the license allows, and what the maker charges
thecontinuitylayer.com/open-models/gemma-4Google DeepMind's family of open models in five sizes (E2B, E4B, 12B, 26B A4B and 31B), with a 256K context window on the medium models and Apache 2.0 licensing; the two largest also run on the Gemini API.
- 1Yes
Is Gemma 4 under an open license?
Yes, by Google DeepMind's own page.
The Hugging Face card lists the license as Apache 2.0 and links Google's Gemma 4 license page, which carries the Apache License 2.0 text with no added conditions; Google's Gemma docs say the models permit responsible commercial use.
Yes means: The weights are under an OSI-approved license such as Apache-2.0 or MIT. Partly means the maker's own license with conditions, such as a user cap or an acceptable-use clause; no means research or non-commercial use only.
- 2Yes
Can you download Gemma 4's weights?
Yes, by Google DeepMind's own page.
Weights for all five sizes, base and instruction-tuned, are on Hugging Face under google in BF16, with a quantizations list for llama.cpp, Ollama and LM Studio; the card shows no access request step.
Yes means: The maker's page links the weights for download with nothing more than a click-through license. Partly means an access form, an approval step, or only some sizes released.
- 3Yes
Does the maker state Gemma 4's context window?
Yes, by Google DeepMind's own page.
The card says "Gemma 4 features a context window of up to 256K tokens": 256K on the 12B, 26B A4B and 31B models and 128K on E2B and E4B.
Yes means: A token count for the context window is on the maker's page or model card.
- 4Yes
Does the maker say what it takes to run Gemma 4 on your own machine?
Yes, by Google DeepMind's own page.
Google's Gemma docs give the memory to load each size's weights: 31B needs 69.9GB in BF16, 34.9GB at 8-bit and 17.5GB at 4-bit; 26B A4B 57.7, 28.8 and 14.4GB; 12B 26.7, 13.4 and 6.7GB; E4B 17.9, 8.9 and 4.5GB; E2B 11.4, 5.7 and 2.9GB. The figures exclude the context window.
Yes means: The page states the parameter count and the memory or hardware it needs. Partly means the parameter count alone.
- 5Partly
Does the maker publish a price for its hosted Gemma 4?
Partly, by Google DeepMind's own page.
Google's Gemma docs say the Gemini API supports gemma-4-31b-it and gemma-4-26b-a4b-it with an API key from AI Studio; the page gives no price, and the Gemini API pricing page lists no Gemma model.
Yes means: A per-token price on the maker's own pricing page. Partly means a hosted version exists but the price sits behind sign-in or a sales call.
- 6Partly
Does the maker train on your prompts when you use its hosted Gemma 4?
Partly, by Google DeepMind's own page.
The Gemini API pricing page's tier overview says Free: "Content used to improve our products" and Paid: "Content not used to improve our products"; which tier a Gemma call falls under is not stated on the page.
Yes means: Here no is the good answer: the maker's page says what you send its API is not used for training. Yes means it is unless you opt out; partly means it depends on the plan or the page does not give the default.
- 7Yes
Can you fine-tune Gemma 4 and keep the result?
Yes, by Google DeepMind's own page.
Apache 2.0 per the card, which states no condition on fine-tuning, derivatives or redistribution and lists downstream fine-tunes, adapters and merges in its model tree.
Yes means: The license allows fine-tuning and the fine-tuned weights are yours to keep and share. Partly means conditions follow the derivative, or a hosted fine-tune you cannot download.
License as stated: Apache 2.0; the Hugging Face card links Google's Gemma 4 license page, which carries the Apache License 2.0 text.. Every grade above is read from Google DeepMind's own page on the day shown. If a page has changed, tell us.