Google Releases Gemma 4 with Native Vision and 256K Context
2026-04-09
“Google quietly dropped the first open model family that doesn't feel like a consolation prize. Four sizes, native multimodal, 256K context, 140 languages. Meta is watching this through the bars of the cage they built themselves.”
Gemma 4 is the first Google model release where the word 'finally' isn't doing any heavy lifting. Native vision, native audio, native function calling, 256K context, and a 140-language footprint that makes the previous generation look like it was trained in a single time zone. All four sizes ship under a license that lets you actually use them.
The timing is surgical. Meta announced two weeks ago that LLaMA is going closed after LLaMA 5. Google looked at that, looked at its own history of half-committing to open source, and decided the one thing that would hurt Meta the most was to out-open them in the same news cycle. And it worked.
The benchmarks are in the range where 'benchmark' stops meaning much. Real test: can a solo developer in a bedroom download the 9B parameter version and ship something that doesn't embarrass itself? Yes. That's the whole game now.
- Gemma 4 released in four sizes (2B, 9B, 27B, 90B) with native vision, audio, and function calling built in from day one.
- Context window expanded to 256K tokens, matching the frontier closed models.
- Support for 140+ languages out of the box, up from 35 in Gemma 3.
- License permits commercial use and model distillation, directly contradicting Meta's new closed-source pivot.
- 01
For the first time in years, a solo developer with a laptop and no OpenAI API key can ship a legitimately competitive product. The gap between 'what big labs can do' and 'what I can do in my bedroom' just shrank by a generation.
- 01
Meta, which announced two weeks ago that LLaMA 5 would be closed source 'for safety reasons'. Google just made the safety argument look like an excuse for falling behind.
No meter. No surprises.
The story isn't Copilot. It's the meter. Here is the calmer way to get your code reviewed.
- Flat price for the Full Suite. No usage meter, no month-end surprise.
- Free CLI: 90 roasts a month, no account needed.
- Privacy-first: a dry-run shows the exact payload, and your secrets never leave your machine.
Works in Claude Code, Cursor and Windsurf via MCP. Open source, and proud of it.
Got something the world should see roasted? Drop it.
A full teardown from €2,99. No mercy.