Gemini 4 Carbon

As of November 9, 2026, Gemini 4 Carbon by Google costs $2.00 per 1M input tokens and $10.00 per 1M output tokens, with a 1M-token context window.

Gemini 4 Carbon is the model Google calls its strongest, and it reached developers the same way Gemini 4 Argon did: restricted first, priced before anyone could buy it. Google DeepMind announced it on October 28, 2026 and gave it first to a subset of cyber defenders and then trusted partners in its Fairwind Program. The Gemini API changelog records general availability on November 10, 2026 as gemini-4-carbon, on the paid tier through Google AI Studio, Android Studio, and Gemini for business, while Google AI Ultra subscribers get it in the Gemini app, AI Mode in Search, and Antigravity. It is not on the Free tier. The rate card matches Argon line for line: an introductory $2 per million input tokens and $10 output through November 30, 2026, rising to $4 and $20 after that, with context caching at $0.10 and then $0.20, implicit only. The 1,048,576 token window is unified across input, thinking, and output, so a long prompt and a long answer compete for the same budget; default synchronous output is 65,536 tokens, and long decode raises the ceiling to the full window. Input spans text, image, video, audio, and PDF, output is text only, thinking runs at low, medium, or high, and the model card gives a March 2026 knowledge cutoff. Google does not pitch Carbon as the default. Its launch post keeps Argon as the right pick for most software engineering and knowledge work and says Carbon spends more tokens reasoning and checking its own work on harder tasks, so the same job can cost more on Carbon at the same per-token price. Google reports first place on CursorBench 4.0, Perplexity's WANDR, and Zapier's AutomationBench at more than 2x lower cost per task than the next best external model. Independent measurement at announcement backs the headline: as of October 28, 2026 Artificial Analysis ranked Carbon first on its Intelligence Index v5.0 at 59, one point above Claude Opus 5.5 at 58, and Vals measured 63.13 percent on Terminal-Bench 4.0, fourth behind Opus 5.5 at 65.15. Anthropic's Claude Fable 5.2 launch table, published the day before GA, shows Fable 5.2 ahead on seven of the eight rows that include Carbon, with Carbon leading only AutomationBench, 53.9 percent to 51.4.