Google’s Gemini 4 Carbon targets deeper reasoning and coding
Google has made Gemini 4 Carbon available on the paid tier of the Gemini API and in Google AI Studio, positioning it as its strongest reasoning model. It targets sustained agent workflows across software engineering, multimodal knowledge work, scientific reasoning, and cybersecurity defense, spending more thinking tokens to explore hypotheses and verify results.
Carbon shares Gemini 4 Argon’s unified 1M-token context window, built-in tools, and per-token pricing, but uses more tokens per turn and takes longer on complex tasks. Google highlights deeper build-and-verify loops, stronger source grounding, and stricter output-schema adherence. Developers can choose low, medium, or high thinking levels, with medium as the default.
Standard synchronous responses stop at 65,536 tokens; longer generation requires background execution and a higher output limit within the shared context window. Carbon also works with the Antigravity agent in Gemini Managed Agents. Migration from Argon requires no request-schema changes, while older integrations must remove explicit context caching and account for ignored sampling parameters.
Introductory pricing runs through November 30, 2026, at $2.00 per 1M tokens for input and $10.00 per 1M tokens for output, including thinking tokens. Those rates rise to $4.00 and $20.00, respectively, starting December 1, 2026. Every processed token is billed, including internal tool instructions and function declarations when grounding is used.