Google DeepMind announced the general availability of Gemini 4 Carbon on November 10, 2026. Carbon is the company's strongest model, aimed at long agentic workflows in knowledge work, software engineering and scientific reasoning. It sits next to Gemini 4 Argon, which Google launched earlier.
Who gets it and what it costs
All paying developers and enterprises can call Carbon through the API in Google AI Studio and in Gemini for business. The introductory price equals Argon's: $2 per million input tokens and $10 per million output tokens. A footnote in the announcement says the introductory period ends on November 30, 2026. After that the price becomes $4 per million input tokens and $20 per million output tokens. The footnote also states that these terms apply to the publicly available Carbon model.
Google AI Ultra subscribers get Carbon in Google Antigravity, the Gemini app and AI Mode in Search. Scientists can apply for free access to Argon and Carbon for six months through the Antigravity program, with onboarding on a rolling basis. A separate version, Carbon for cyber, goes only to defenders accepted into the Fairwind Program.
Carbon or Argon
Google positions Argon as the default for most software engineering and knowledge work. Carbon is meant for tasks that push a model hardest, such as complex multi-step coding and debugging and open-ended research. It uses more tokens than Argon, because it explores more, reasons longer and runs extended verification loops. Higher token use matters when the per-token price is about to double.
Benchmarks Google cites
Google says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench, at more than twice as low a cost per task as the next best external model. On the new Multimodal Knowledge Agent benchmark, built with Artificial Analysis for agents working over dense files with charts, tables and schematics, Carbon ranks second at more than six times lower average cost per task than the top model. These are Google's reported results, and the cost comparisons depend on the introductory pricing.
Agents in Antigravity
Google is previewing a /swarm feature in Antigravity. It turns one prompt into a coordinated team of agents that can be tracked in real time. Developers can give Carbon the roles that need deep reasoning, such as design, coordination and auditing, and pair it with lighter, faster models in parallel workspaces, with adversarial verifiers testing each change.
Safety claims
Google reports a per-attempt breach rate of 0.8% in simulated cyberattack tests, down from 3.1% on Argon, and 92% of tasks passed on its 375-task agentic honesty benchmark, against 70% for Argon. It also says Carbon reached a 0% attack success rate on the Gray Swan indirect prompt injection benchmark, and admits that prompt injection robustness still needs work and that some sophisticated attacks could do better. Those figures come from Google and its testing partners.
Google's updated Gemini 4 developer guide covers migrating existing workflows to the model.
Top comments (0)