Google releases Gemini 4 Carbon to the public after cybersecurity preview
Google has opened Gemini 4 Carbon to the general public two weeks after a limited release to a select group of cybersecurity defenders. The company calls Carbon its strongest model yet, able to compete with the leading systems from Anthropic and OpenAI, and says it pushes the frontier further than Argon, the other Gemini frontier model that launched only last month. Argon is still expected to remain the better default for most software engineering and knowledge work; Carbon is aimed at work that tests a model’s limits, including complex multi-step coding and open-ended scientific research.
To underline Carbon’s use on long-horizon agent tasks, Google is previewing a swarm feature on its Antigravity agent development platform, which can turn a single prompt into a coordinated team of agents. Carbon is available today through the Gemini API in Google AI Studio and Gemini for business for all paying developers and enterprises, and to all Google AI Ultra subscribers in the Gemini app, AI Mode in Search, and Antigravity. It is not clear when, or whether, it will reach lower-tier or free users.
Introductory pricing is $2 per million input tokens and $10 per million output tokens — about a fifth the cost of OpenAI’s GPT-6 Astra and a quarter that of Anthropic’s Fable 5.2, which launched yesterday. Those rates are due to double at the end of November, after which Carbon should still be cheaper than rivals on a per-token basis, though models can consume different numbers of tokens for the same job. Google’s cited benchmarks place Carbon among the most powerful models on most metrics, behind Fable 5.2; such scores are often cherry-picked. Coming so soon after Argon, the launch is presented as Google’s return to the front of AI research after months in which that race looked like a contest between Anthropic and OpenAI.
Continue in the app — vote & join in ➔