Google launches Gemini 4 Carbon API with pricing

Google launched Gemini 4 Carbon on Nov. 10, making the model available through the Gemini API in Google AI Studio and Gemini for business for paying developers and enterprises. Google said the model is also rolling out to Google AI Ultra subscribers in Google Antigravity, the Gemini app, and AI Mode in Search.

The release gives builders a new frontier model with a 1 million-token context window and support for low, medium default, and high thinking settings. Google AI Studio’s latest models page identifies the model ID as gemini-4-carbon and says the model is designed for deliberate speed, with a 64K default output limit or up to 1M output tokens with long decode.

What is Google charging for Gemini 4 Carbon?

Google said introductory API pricing is $2 per 1 million input tokens and $10 per 1 million output tokens through Nov. 30, 2026, then rises to $4 per 1 million input tokens and $20 per 1 million output tokens starting Dec. 1, 2026. Google said the introductory public-access API price matches Gemini 4 Argon’s launch pricing, and its documentation says the same post-introductory pricing applies to the publicly accessible Carbon model.

For founders and AI teams, the pricing structure matters because Google says every token processed by Gemini 4 Carbon is billed, including thinking tokens and certain internal tool instructions and function declarations used with grounding. Google AI Studio also says the model is not available on the Free tier.

The company is also charging separately for higher-throughput options. Google AI Studio lists Batch input at $1.00 per 1 million tokens through Nov. 30, 2026 and $2.00 starting Dec. 1, 2026, while Priority input is priced at $3.60 and then $7.20. Batch output, including thinking tokens, is listed at $5.00 and then $10.00, while Priority output is $18.00 and then $36.00.

How does Carbon fit into Google’s model lineup?

Google said Argon remains the better fit for most software engineering and knowledge work, while it positioned Carbon for complex multi-step coding, debugging, open-ended scientific research and long-horizon agentic workflows. In Google’s telling, Carbon uses more tokens than Argon because it spends more time on deeper exploration, reasoning and extended verification loops.

Google AI Studio’s Carbon guide says the model is built for sustained agentic workflows across complex software engineering, multimodal knowledge work, scientific reasoning and cybersecurity defense. The guide also says Carbon’s default thinking level is medium and that it spends more thinking tokens on hard problems to explore hypotheses, verify work and check edge cases before responding.

Google said Carbon is its strongest model for cybersecurity capabilities and long-horizon agentic workflows across complex knowledge work, software engineering and scientific reasoning. The company said it had released Carbon to cybersecurity defenders through its Fairwind Program two weeks before the public availability announcement.

What does Google say about performance and safety?

Google paired the launch with a long list of benchmark claims. It said Carbon ranked No. 1 on CursorBench 4.0, Perplexity’s WANDR and Zapier’s AutomationBench, while costing more than 2x less per task than the next best external model. It also said a new Multimodal Knowledge Agent benchmark from Google and Artificial Analysis placed Carbon No. 2 while costing more than 6x less per task than the top-performing model.

On research and science use cases, Google said Carbon matched the best tested model on Hebbia’s deep financial research evaluation and ranked at the top of 23 frontier model configurations on Balyasny Asset Management’s benchmark. The company also said Axiom Bio found Carbon to be its leading model for translational science and is moving production scientific analyses and agents to Carbon and Argon.

Google highlighted cybersecurity testing as well. It said Carbon led on key cyber benchmarks through the Fairwind Program across vulnerability finding, exploiting and patching; reached 100% recall on scanning and 95% precision on triage in Project Fortify’s internal benchmark; blocked 99% of attack trajectories in simulated cyberattack testing; and cut the per-attempt breach rate to 0.8% from 3.1% on Argon. Google also said Gray Swan testing lowered automated cyberattack query success rates to 2.17% from 9.86% on Argon, and that external testing by Trajectory Labs found no new universal jailbreaks across 132 hours and more than 13,000 requests. These are Google-reported results and were not independently verified in the supplied evidence.

What happens next for developers and scientists?

For developers, the practical next step is access through Google AI Studio and Gemini for business, where paying users can start building with the model ID gemini-4-carbon and its 1 million-token window. Google AI Studio says explicit context caching is not supported, but implicit caching is enabled by default for prompts of at least 4,096 tokens with a 95% discount on cache hits. The documentation also says SDKs will automatically continue Carbon generations across requests by default when max_output_tokens exceeds per-request caps.

Google is also expanding access on the consumer and research side. It said Carbon is rolling out to Google AI Ultra subscribers in Google Antigravity, the Gemini app and AI Mode in Search, and that it is expanding access for current and future participants in its scientist program. Applications for that program remain open, with weekly rolling onboarding subject to eligibility, capacity and selection.

Based on Google’s product description and pricing, the launch suggests Google is positioning Carbon as a heavier-duty model for tasks that need sustained reasoning, verification and multi-step coordination. The company said developers can use it for architectural design, coordination and auditing while pairing it with faster models in parallel isolated workspaces and adversarial verifiers.