Wednesday, 11 November 2026

Gemini 4 Carbon opens to paying developers and Ultra subscribers

Google's strongest model is now in the Gemini API at $2 per million input tokens, and in the Gemini app for AI Ultra subscribers.

Key points

  • Gemini 4 Carbon is now in the Gemini API for paying developers at $2 per million input tokens.
  • AI Ultra subscribers get it in the Gemini app, AI Mode in Search and Antigravity.
  • Google is previewing a /swarm feature that turns one prompt into a team of coding agents.

Google has made Gemini 4 Carbon, its most capable model so far, available outside the security community. Paying developers can call it through the Gemini API from today, and people on the AI Ultra plan can pick it in the Gemini app and in AI Mode in Search.

Carbon first went out two weeks ago, but only to cybersecurity teams. Google is charging its launch rate: $2 for every million tokens you send in and $10 for every million that come back, identical to what Gemini 4 Argon costs. Google positions Carbon for the hardest jobs, such as multi-step debugging and open-ended research, and says it spends more tokens than Argon thinking a problem through. Argon stays the recommended pick for most coding and office work.

Google’s announcement says Carbon tops three third-party tests, one from Cursor for coding, one from Perplexity for research agents and one from Zapier for workflow automation, while costing less than half as much per task as the nearest outside rival. These are Google’s own picks of results, so treat them as claims until independent tests appear. On a new multimodal benchmark built with Artificial Analysis, which tests how agents read charts, blueprints and dense PDFs, Carbon placed second, at about a sixth of the average cost per task of the leader.

Alongside the model, Google is previewing a /swarm command in Antigravity, its agent-based coding tool. It turns one prompt into a team of agents that work in separate workspaces, with checkers that try to break each change. In Google’s setup, Carbon handles planning and review while cheaper, faster models take the bulk of the work.

On safety, Google reports that in simulated attack tests Carbon blocked 99% of the red-team agent’s attack paths, and its per-attempt breach rate fell to 0.8% from 3.1% for Argon. Google also says the model passed 92% of tasks on its 375-task honesty test, against 70% for Argon. The company adds that resistance to prompt injection, where hidden text in a page or file steers an agent, still needs work.

The release follows Carbon’s October debut for defenders. Defenders can still apply to the Fairwind programme. Researchers get a boost too: the scheme that gave 10,000 scientists six months of free Antigravity access with Argon now covers Carbon as well.

Google’s post names only Ultra subscribers for the Gemini app and Search, so there is no stated date for cheaper plans or free users. Developers on the free API tier are also not listed; the paid tier of Google AI Studio is the route in.

TPRO Team

The Tech Prolonged newsroom desk. Stories under this byline cover AI, the web and general tech, and are reviewed by our editors before publishing.