Gemini 4 Carbon Opens to All Paying Developers, and Its Price Doubles on December 1

On November 10, 2026, Google made Gemini 4 Carbon available beyond the cyber defenders who received it first on October 28. A post by Koray Kavukcuoglu says it is now on the API "for all paying developers and enterprises" through Google AI Studio and Gemini for business, and is "rolling out to all Google AI Ultra subscribers" in Google Antigravity, the Gemini app and AI Mode in Search. The Gemini API changelog lists it as generally available under the model ID gemini-4-carbon. It is not on the free tier.

Google positions Carbon as the deeper, slower sibling of Gemini 4 Argon. The Gemini API documentation says Carbon "spends more thinking tokens per turn to explore thoroughly and verify its work, which increases per-turn token use and latency compared to Argon", and recommends keeping Argon for interactive coding and "fast production loops".

What Google reports

All of these figures come from Google or its launch partners, and we have not reproduced any of them.

Benchmarks. Google says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench "at more than 2X lower cost per task compared to the next best external model".
Misuse safeguards. In simulated cyberattack environments, Google says Carbon "blocked 99% of attack trajectories" from an adaptive red-teaming agent, with a per-attempt breach rate of 0.8 percent against 3.1 percent for Argon. In Gray Swan's testing, automated cyberattack queries succeeded 2.17 percent of the time, against 9.86 percent on Argon.
Honesty. On a 375-task agentic honesty benchmark it passed 92 percent of tasks, against 70 percent for Argon.
Multi-agent work. Google is previewing a /swarm feature in Antigravity that turns one prompt into a team of agents, with Carbon in coordinating and auditing roles and lighter models doing parallel work.

The migration notes that change older code

For teams already on Argon, the "What's new" page says upgrading "requires no request schema changes": change the model ID and allow for longer thinking. For applications coming from Gemini 3 or earlier, it lists changes that will matter more than any benchmark:

Caching. "Explicit caching (cachedContents / caches.create) is not supported on Gemini 4 Carbon." Implicit caching is on by default for prompts of at least 4,096 tokens and discounts cached input by 95 percent.
Sampling. Applications should remove temperature, top_p and top_k. "If provided, they are ignored."
Time and length. Synchronous responses stop at 65,536 tokens, longer output needs background execution, and Google advises HTTP timeouts of "at least 15 minutes" because hard prompts "can take several minutes".

Price applies to everyone. Input costs $2 and output $10 per million tokens "through November 30, 2026", then $4 and $20 from December 1, and thinking tokens are billed as output.

What remains uncertain. The cost-per-task claims depend on how many thinking tokens a task uses, which varies with the workload and the thinking level chosen. The safety figures come from Google's own environments and two partners, and the introductory price makes early cost comparisons flattering for three weeks.

Analysis: Budget at the December price, not the launch price. If Tanishka's team at Belgavi's Book Store tested Carbon on its hardest help-desk cases, the useful measure is cost per resolved CASE at $4 and $20, compared with Argon on the same cases at each thinking level. If that code started on Gemini 3 or earlier, search it for any setting of temperature, because Carbon ignores it, and a test that relied on it for repeatable answers will pass or fail for a different reason. Any explicit cache call also has to go, along with the savings it was providing.

Sources and evidence

Gemini 4 Carbon became generally available to paying API customers and AI Ultra subscribers on November 10, at $2 and $10 per million tokens until November 30 and $4 and $20 after. Its benchmark and safety figures are Google's own. Argon users change only the model ID, but apps from Gemini 3 or earlier lose explicit caching and sampling settings. Test at December prices.