Gemini 4 Carbon opens to all developers at $2 per million input tokens

Google has listed Gemini 4 Carbon (gemini-4-carbon) as generally available in the Gemini API, according to the Gemini API changelog entry dated November 10, 2026. Introductory pricing is $2.00 per million input tokens and $10.00 per million output tokens.

What the changelog entry says

The entry appeared on the Gemini API changelog, which now redirects to the AI Studio docs. It describes Carbon as Google’s most capable reasoning model, built for sustained agentic coding, multimodal knowledge work, scientific reasoning and cybersecurity defense. It links to a model page, a pricing section and a guide titled “What’s new in Gemini 4 Carbon”.

The listed rates are introductory and run through November 30, 2026. The entry does not say what the price will be afterward. It also does not list rate limits, context window size or regional availability, and no benchmark figures appear in the entry itself.

From restricted preview to open API in two weeks

The change matters because of how Carbon first appeared. According to Reuters, as carried by Khaleej Times, Google unveiled the model on October 28, 2026, two weeks after Gemini 4 Argon. At that point it was available only to a subset of cybersecurity defenders through Google’s Fairwind Program, while the company kept testing safeguards. Reuters quoted Google as claiming “state-of-the-art performance across cybersecurity tasks”, including vulnerability discovery, exploitation and software patching.

Late-October coverage of the launch also found no public model ID and no date for wider access. The changelog entry now supplies both: a model ID developers can call and a general availability label. Any developer with Gemini API access can now use the model, according to the entry, though the page does not say whether the Fairwind safeguards carry over to the public version.

Reuters reported that the $2 and $10 rates match the introductory pricing of Gemini 4 Argon. The changelog shows Argon reached general availability on October 13, 2026, so Carbon has gone from announcement to API availability in a shorter window than Argon had between its own milestones. Developers already on Argon can compare the two at identical per-token rates, which makes the choice a question of capability rather than cost.

What to watch before the end of November

The first signal is the price after November 30. If Google keeps the rates, Carbon becomes a straightforward default for coding agents. If it raises them, teams that budget around $2 and $10 will need to revisit that. The second is independent testing. Google’s claim, reported by Reuters, that Carbon generates secure code at more than twice the success rate of Argon comes from internal benchmarks and has not been independently verified.

The third is how Google handles the cyber capabilities now that the audience is wider. A model positioned around exploitation and patching raises the stakes for the safeguards Google said it was still testing on October 28. The changelog entry does not describe any change to those controls, and the story is worth revisiting once the linked guide and model page show what limits apply.

Rivals are also positioning against Google’s lineup. Our report on Perplexity adding Fable 5.2 noted a claimed lead over Gemini Argon, and Devin’s own test, covered in our FrontierCode report, put Fable 5.2 on top at 49.1%. Neither compared against Carbon, so head-to-head results are still missing.

For now, the verified facts are narrow: a model ID, a GA label, two price points and a November 30 date. Everything else, including limits and the safeguards behind the cyber features, depends on documents the changelog only links to.