This post is written in English.
Google is making Gemini 4 Carbon available to the public. The model first went to a select group of "cybersecurity defenders" two weeks before this rollout. Google says Carbon is its "strongest model" yet, and says it can compete with the best systems from Anthropic and OpenAI. Those are Google's claims. The company hasn't published independent results to back them up yet.
Where Carbon fits
Carbon sits above Argon, Google's other frontier model, which launched last month. Google says Carbon "pushes the frontier even further" than Argon. It also says Argon will likely stay "the ideal choice for the majority of software engineering and knowledge work." Carbon is aimed at harder work: Google says it is "built for tasks that test the limits of a model's intelligence," such as complex, multi-step coding and open-ended scientific research.
The swarm feature
Google is previewing a "swarm" feature on Antigravity, its agentic development platform. According to Google, it lets a user turn a single prompt "into a coordinated team of agents." The company is pitching this as a way to use Carbon for long-running agent work. The announcement doesn't say how well the feature performs, and the preview is the only detail given.
Who gets access
Carbon is available now through the Gemini API in Google AI Studio and through Gemini for business, for all paying developers and enterprises. Google AI Ultra subscribers also get it in the Gemini app, in AI Mode in Search, and on Antigravity. It hasn't said when, or whether, lower-tier and free users will get it.
Pricing
The introductory price is $2 per million input tokens and $10 per million output tokens. The Verge reports that this is roughly a fifth of the cost of OpenAI's GPT-6 Astra and a quarter of Anthropic's Fable 5.2, which launched the day before. The introductory period ends at the end of November, and after that the prices double. Carbon would still be much cheaper per token than those rivals at the new rate.
Per-token price doesn't tell you the total cost of a task, though. Models can use different numbers of tokens to do the same job.
Benchmarks
Google's launch benchmarks put Carbon among the most capable models on most measures, behind Fable 5.2. The Verge notes that companies can and do pick benchmark results that show their tools in the best light. These numbers come from Google and haven't been independently verified.
My read
The most useful number here is the price, and it's also the least settled. A low introductory rate is a real offer, but the November increase is part of the deal, so I'd compare costs at the doubled rate. For agent and coding work, the swarm feature and the claims about hard tasks are the things to test yourself, since the only evidence so far is Google's own. Carbon may be a serious option for teams that need frontier capability and are sensitive to per-token cost. Until there are independent results, I'd treat "strongest model" as a marketing claim.
← All posts


