Google's Gemini 4 Carbon is out on the API at $2/$10 until 30 November

Google made Gemini 4 Carbon, its strongest model, generally available on 10 November 2026. Koray Kavukcuoglu, Google DeepMind's SVP, wrote in the launch post that it is now on the Gemini API paid tier in Google AI Studio and in Gemini for business, and is rolling out to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and Google Antigravity.

Google announced Carbon on 28 October and gave it first to cybersecurity defenders through its Fairwind Program. Public availability followed 13 days later, the same gap Argon had between announcement and release.

What it costs

Google prices Carbon at the same introductory rate as Argon: $2 per million input tokens and $10 per million output tokens. A footnote in the post says that after the introductory period expires on 30 November 2026, $4 per million input tokens and $20 per million output tokens apply. That doubling lands three weeks from now, so any pilot budget should use the December rate.

The per-token price is half the story. Google says Carbon uses more tokens than Argon "to explore thoroughly, reason deeply, and run extended verification loops", and that Argon "remains the ideal choice for the majority of software engineering and knowledge work". Google positions Carbon for complex multi-step coding, debugging and open-ended research.

What Google claims

Google says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench at more than twice lower cost per task than the next-best external model. On a new Multimodal Knowledge Agent benchmark that Google built with Artificial Analysis, it says Carbon ranks second at more than six times lower average cost per task than the leader. These are Google's summaries of third-party benchmarks; the post does not give the scores.

On security work, Google says that in its internal Project Fortify system Carbon hit 100% recall on scanning and 95% precision on triage, on Google's own benchmark. It also says Wiz used Carbon to find zero-day vulnerabilities at two large SaaS providers in under two hours.

On safeguards, Google reports that Carbon passed 92% of tasks on its 375-task agentic honesty benchmark, against 70% for Argon. It says Gray Swan's testing put the automated cyberattack query success rate at 2.17%, against 9.86% for Argon, and that Trajectory Labs found no new universal jailbreaks in 132 hours and over 13,000 requests.

Agent swarms in Antigravity

Google is also previewing a /swarm feature in Antigravity that turns one prompt into a coordinated team of agents. Google suggests assigning Carbon to roles such as architecture, coordination and auditing, and pairing it with faster, lighter models in parallel workspaces with adversarial verifiers. The post gives no pricing or quota for swarm runs.

What the post leaves out

The post gives no context-window size, rate limits or per-task token counts, and none of the benchmark scores behind Google's rankings. It also says the consumer rollout is to Google AI Ultra subscribers, so teams on other plans should not expect Carbon in the Gemini app yet.

For a team already on Argon, the API route is the one Google says is open to all paying developers today. Run your own tasks on both models, record tokens per completed task, and price the result at $4 and $20.