The short version
Gemini 4 Carbon is now open to every paying Gemini API customer, at Argon-level introductory prices that double after November 30. Google positions it for the hardest coding and research tasks and keeps Argon as the default for most work. Teams doing generative AI integration should add Carbon as a routed option, not a blanket replacement.
What did Google release on November 10?
Google moved Gemini 4 Carbon from a restricted preview to general paid access. The model was announced on October 28 and first went to a group of security partners while Google ran pre-release testing under the voluntary US government framework, Ars Technica reported at the time. On November 10 Google opened it to all paying developers and enterprises through the Gemini API and Gemini for business, and to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and Antigravity.
The Verge reports it is not clear when, or whether, Carbon will reach lower subscription tiers or free users.
How is Carbon different from Gemini 4 Argon?
Google does not present Carbon as a replacement for Argon. In its words, Argon will likely remain “the ideal choice for the majority of software engineering and knowledge work,” while Carbon is built for tasks that test the limits of a model, such as complex multi-step coding and open-ended scientific research. Google also says Carbon uses extended reasoning with longer token usage than Argon.
| Claim | What the sources report |
|---|---|
| Agent benchmarks | Google: first place on CursorBench 4.0, Perplexity’s WANDR and Zapier’s AutomationBench, at more than 2× lower cost per task than the next best external model |
| Agentic honesty | Google: 92% of tasks passed, versus 70% for Argon |
| Prompt injection | Google: 0% attack success on the Gray Swan indirect prompt injection benchmark |
| Secure code | Ars Technica, citing Google: about twice as likely as Argon to produce code without vulnerabilities |
| Size of the lead | Ars Technica: the gap between Carbon and rival frontier models is only a few points in most tests |
How much does Gemini 4 Carbon cost?
Carbon launches at $2 per million input tokens and $10 per million output tokens, the same introductory rate as Argon. From December 1 the price doubles to $4 and $20. The Verge calculates the launch price at roughly a fifth of OpenAI’s GPT-6 Astra and a quarter of Anthropic’s Fable 5.2, which launched on November 9.
The same report adds an important caveat: per-token prices do not translate directly into per-task costs, because models spend different numbers of tokens on the same job. With Carbon reasoning longer than Argon, a cheaper rate card can still produce a bigger bill on some workloads.
What is the swarm feature in Antigravity?
Swarm is a preview capability in Antigravity, Google’s agentic development platform. Google says it turns a single prompt into a coordinated team of agents for long-horizon projects.
It is also the part that needs the most care: several agents working in parallel multiply token spend, widen access to code and tools, and make it harder to trace which agent made which change.
What it means for US & EU software teams
Hard-coding one model into a product is now a liability. Frontier models ship every few weeks: Anthropic released Fable 5.2 on November 9, and Google followed with Carbon a day later. The more durable pattern is a routing layer with three parts:
- a cheaper default model for most requests;
- a frontier model for the tasks that fail;
- an evaluation set that decides which is which.
The November 30 price change matters for budgets. A proof of concept priced at the introductory rate will cost about twice as much per token in December. If you plan a pilot, model the cost at $4 / $20, not $2 / $10, and measure tokens per completed task rather than price per million tokens.
Google’s security numbers are encouraging, especially the prompt injection result, but they come from vendor testing. Agents that read untrusted content, call tools or write to repositories still need sandboxing, least-privilege credentials and human review before merge.
For EU teams, check where Gemini for business processes and stores data before sending customer or regulated data to a new model.
What should teams do this month?
- Keep your default. Leave Argon or your current model in place for routine coding and content tasks.
- Collect the failures. Pull the prompts where your current model fails or needs several retries, and run them on Carbon.
- Measure cost per task. Compare tokens used and success rate, not list price, and use the post-November 30 rates.
- Trial swarm in a sandbox. Give multi-agent runs a separate repository, scoped credentials and a spending cap.
- Make switching cheap. Put models behind one interface so the next release is a config change, not a rewrite.
Frequently asked questions
When did Gemini 4 Carbon become publicly available?
Google rolled Gemini 4 Carbon out on November 10, 2026, two weeks after it was announced on October 28 and released first to a group of cybersecurity defenders. It is available through the paid Gemini API in Google AI Studio, in Gemini for business, and to Google AI Ultra subscribers in the Gemini app, AI Mode in Search and Antigravity.
How much does the Gemini 4 Carbon API cost?
Introductory pricing is $2 per million input tokens and $10 per million output tokens until November 30, 2026. After that, Google lists $4 per million input tokens and $20 per million output tokens. The Verge notes that a lower per-token price does not always mean a lower cost per task, because models use different numbers of tokens for the same job.
Should teams switch from Gemini 4 Argon to Carbon?
Not by default. Google itself says Argon will likely remain the ideal choice for most software engineering and knowledge work, while Carbon is built for the hardest tasks, such as complex multi-step coding and open-ended research. Google also says Carbon reasons longer than Argon, so it can use more tokens per request. Route only the tasks that fail on Argon to Carbon and measure the result.
What is the swarm feature in Google Antigravity?
Swarm is a preview feature in Antigravity, Google's agentic development platform. According to Google, it turns a single prompt into a coordinated team of agents for long, complex projects. It is in preview, so teams should treat it as an experiment rather than a production workflow.
Are Google's Gemini 4 Carbon benchmarks reliable?
They are vendor-reported. Google says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench. Ars Technica noted that the gap between Carbon and rival frontier models is only a few points, and The Verge warned that companies can cherry-pick benchmarks. Teams should test Carbon on their own tasks before changing defaults.
Sources
Google — Gemini 4 Carbon is now available (primary source, November 10, 2026)
The Verge — Google releases Gemini 4 Carbon with agent ‘swarm’ feature (November 10, 2026)
Ars Technica — Google announces Gemini 4 Carbon, its latest frontier AI model (October 28, 2026)