Google makes Gemini 4 Carbon, its strongest model, generally available
Two weeks after a release limited to cybersecurity defenders, Carbon is in the API for paying developers at $2 per million input tokens and $10 per million output tokens, and rolling out to AI Ultra subscribers.
Why it matters
Carbon arrives a day after Anthropic's Fable 5.2, which Anthropic's own table placed ahead of Carbon on most tests, so the two newest frontier models are now on sale at once. Google pitches Carbon at the hardest work, multi-step coding and open-ended research, and says its lighter Argon model remains the better choice for most tasks. It claims first place on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench at more than twice lower cost per task than the next best external model. Watch independent benchmark results and how pricing changes after the introductory period.
Key facts
- Available in the API for all paying developers and enterprises
- $2 per million input tokens, $10 per million output (introductory)
- Rolling out to Google AI Ultra in Antigravity, the Gemini app and AI Mode in Search
- Released to cybersecurity defenders via the Fairwind Program two weeks ago
- Google says Argon remains the better choice for most tasks
- Previewing /swarm multi-agent mode in Google Antigravity