- Google officially launches “Gemini 4 Carbon,” the most advanced AI model from the new-generation Google AI model family “Gemini 4,” for developers, enterprises, scientists, and general users.
- Designed for high-difficulty tasks that test the limits of model intelligence, such as complex multi-step coding, debugging, and scientific discovery.
- In the “Gemini” app, “Gemini 4 Carbon” (“4 Carbon”) replaces the long-standing “3.1 Pro.”
On Tuesday, November 10, 2026 local time, Google officially launched “Gemini 4 Carbon,” the cutting-edge AI model from the new-generation Google AI model family “Gemini 4” announced in late October 2026, making it available to developers, enterprises, scientists, and general users.
“Gemini 4 Carbon” is the most powerful frontier model in Google’s history and a higher-tier model than “Gemini 4 Argon.” While “Gemini 4 Argon,” which has already been rolled out generally for most software engineering and knowledge work, is suitable for those areas, “Gemini 4 Carbon” is designed for highly challenging tasks that test the limits of model intelligence, such as complex multi-step coding, debugging, and scientific discovery. A major feature is its ability to consume more tokens to perform exhaustive searches, deep reasoning, and extended verification loops.
“Gemini 4 Carbon” achieved 1st place in “CursorBench 4.0,” Perplexity’s “WANDR,” and Zapier’s “AutomationBench,” delivering a cost per task more than twice as low as the runner-up external model. Furthermore, it secured 2nd place in the “Multimodal Knowledge Agent (MKA)” benchmark—a multimodal intelligence agent benchmark co-developed by Google and Artificial Analysis—achieving an average cost over 6 times lower than top models.
Additionally, a preview of a new feature called “/swarm” has been launched for the development environment “Google Antigravity,” allowing users to assemble autonomous agent teams from a single prompt with real-time tracking. While assigning roles that require deep reasoning, such as architecture design and auditing, to “Gemini 4 Carbon,” users can build robust systems by running it in parallel with lightweight models.
“Gemini 4 Carbon” is now available through “Google AI Studio” and “Gemini for business” for developers and corporate users utilizing the paid tier of the “Gemini API” at an introductory price ($2 per 1 million input tokens / $10 per 1 million output tokens). Note that following the end of the introductory period on Monday, November 30, 2026, the pricing will be $4 per 1 million input tokens / $20 per 1 million output tokens.
For general users, rollout has begun sequentially for subscribers of the “Google AI Ultra” membership via “Google Antigravity,” the “Gemini” app, and Google Search’s “AI Mode.” Incidentally, in the “Gemini” app, “Gemini 4 Carbon” (“4 Carbon”) is provided as a replacement for the long-standing “3.1 Pro.”
Source: Google (https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/)