Gemma
Verdict
Evidence: Editorial assessmentGemma is Google DeepMind's family of open-weight models. This page used to cover only Gemma 3n, the mobile-first variant; Gemma 4 is now the newest generation. Gemma suits teams that want small, efficient models from a major lab. Check the model card's terms before commercial use.
Quick facts
| Gemma | |
|---|---|
| Maker | Google DeepMind |
| Category | AI Models & Open-Weight LLMs |
| Pricing model | Open weights |
| Status | Superseded |
| Official site | ai.google.dev |
| Data last checked | 5 October 2026 |
Key facts and sources
| Claim | Detail | Evidence | Source |
|---|---|---|---|
| Model family and library licence | Google DeepMind describes Gemma as a family of open-weights LLMs based on Gemini research. The gemma library supports Gemma 2, 3, 3n and 4 and is licensed Apache-2.0. | Verified | google-deepmind/gemma on GitHub, 5 October 2026 |
| Notable releases, July to September 2026 | Reported: Kimi K3 open weights (27 July), DeepSeek V4-Flash (31 July) and V4.1 Flash under MIT licence (10 September), GLM-5.3 under MIT (August), Qwen3.8-Max (API only), and a Gemma 4 update (16 July). | Reported | ThursdAI release log, July 2026, 5 October 2026 |
Pricing
No prices are shown because we have not confirmed them on Gemma's own pricing page. Go to Gemma
Who should use it
- On-device and small-server deployments
- Teams that want a Google-backed open-weight model
Who should skip it
- You need the largest open models (compare Qwen3.8 or Kimi K3)
Limitations and risks
- Weight terms are set per model card and differ from the Apache-2.0 code library.
How it compares: Gemma alternatives
Qwen
Alibaba's Qwen family of language models; Qwen3.8 was released as open weights in August 2026, including a 2.4T-parameter flagship.
gpt-oss
OpenAI's open-weight reasoning models, gpt-oss-120b and gpt-oss-20b, released under the Apache 2.0 licence.
Kimi
Moonshot AI's Kimi models; Kimi K3 is a 2.8T-parameter open-weight model with native vision and a 1M-token context window.
DeepSeek
Chinese AI lab DeepSeek's open-weight language models, released under the MIT licence, plus a chat app and API.
Llama 2
Meta's 2023 open-weight language model family. It is a legacy release, superseded by newer Llama and other open-weight models.
Rating
Not yet rated. We show a rating only when at least four of six criteria have documented evidence. How we rate AI tools
Common questions
What is the latest Gemma version?
Google DeepMind's gemma library supports Gemma 2, 3, 3n and 4, with Gemma 4 the newest generation listed when we checked on 5 October 2026.
Is Gemma 3n still supported?
Yes, the gemma library lists 3n among supported versions. For new projects compare it with Gemma 4.
Sources
- google-deepmind/gemma on GitHub · checked 5 October 2026
- ThursdAI release log, July 2026 · checked 5 October 2026
- Gemma official site