
Gemini 4 Carbon, Google's most capable AI model, became available on 10 November 2026 to every paying developer and enterprise through the Gemini API, Google AI Studio and Gemini for business. It costs the same introductory $2 per million input tokens and $10 per million output tokens as Gemini 4 Argon until 30 November 2026, then $4 and $20. If your business software calls Gemini, now is the time to test it and to check what the price change does to your bill.
What happened
In a post on the Google blog titled "Gemini 4 Carbon is now available", dated 10 November 2026, Koray Kavukcuoglu, Chief AI Architect at Google, said developers, enterprises, scientists and consumers can now try the model. Google calls Carbon its strongest model, built for cyber security work and for long-horizon agentic workflows (tasks where an AI agent works through many steps on its own) in knowledge work, software engineering and scientific reasoning.
This is the second step of a staged launch. When Google first introduced Gemini 4 Carbon on 28 October 2026, access went only to a subset of cyber defenders through its Fairwind Program, and Google said availability for developers, businesses and consumers was coming soon.
Google is not retiring its other model. It says Gemini 4 Argon remains the ideal choice for the majority of software engineering and knowledge work, while Carbon uses more tokens than Argon to explore, reason deeply and run extended verification loops. Carbon thinks for longer, and you pay for that thinking.
Alongside the model, Google is previewing a feature called /swarm in Google Antigravity, which developers use as an app, a command-line tool and code editor extensions. One prompt becomes a coordinated team of AI agents that you can follow in real time, with Carbon in the roles that need the deepest reasoning, such as design and auditing, and faster, lighter models doing the rest.
Key details
| Item | What Google says |
|---|---|
| Who can use it now | All paying developers and enterprises, via the Gemini API in Google AI Studio and Gemini for business |
| Consumer access | Rolling out to all Google AI Ultra subscribers in Google Antigravity, the Gemini app and AI Mode in Search |
| Free tier | Not available on the Free tier of the Gemini API |
| Price until 30 November 2026 | $2 per million input tokens, $10 per million output tokens (thinking tokens count as output) |
| Price from 1 December 2026 | $4 per million input tokens, $20 per million output tokens |
| Batch price until 30 November 2026 | $1 per million input tokens, $5 per million output tokens |
| Context window | Input, thinking and output share one 1,048,576-token window |
| Model name in code | gemini-4-carbon, with thinking levels low, medium (the default) and high |
| /swarm | A preview in Google Antigravity that runs several coordinated agents from one prompt |
The figures come from Google's announcement and from the Gemini Developer API pricing page. A token is a small chunk of text, roughly a word or part of a word, and it is the unit AI providers charge by.
Google also published its own test results. It says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench at more than 2X lower cost per task than the strongest outside model it compared against, and that it came top out of 23 frontier model configurations on Balyasny's financial evaluation. On an honesty test, Carbon passed 92% of tasks, against 70% for Argon. These are Google's own figures.
On safety, Google says Carbon blocked 99% of attack attempts from its own red-teaming agent and that its safeguards cut the breach rate per attempt to 0.8%, from 3.1% on Argon. It adds that Trajectory Labs, an outside tester, found no new universal jailbreaks across 132 hours and over 13,000 requests. Google also says plainly that resistance to prompt injection, where hidden instructions in a document or web page try to hijack the model, "remains an area for continued improvement".
Why it matters
Most small and medium businesses never pick an AI model themselves. They use one through software: a chatbot, a document tool or a custom system a developer built. When a provider changes its models and prices, those decisions end up on your invoice.
Three things stand out. First, the price. The introductory rate ends on 30 November 2026, and from 1 December 2026 both input and output prices double. A workload that is cheap to trial in November will cost twice as much per token in December.
Second, token use. Google's developer notes for Gemini 4 Carbon say it increases per-turn token use and latency compared to Argon, and that every token the model processes is billed, including thinking tokens. The same price per token can still mean a bigger bill per task.
Third, agents. Google's /swarm preview shows where the big labs are heading: several AI agents splitting up a job and checking each other's work. The Antigravity team itself warns that parallel multi-agent systems consume quota much faster than a standard chat. We wrote about AI agents' permissions, data access and cost for a small business; this release makes those questions more pressing.
What this means for businesses
You do not need to move to Carbon because it is new. Carbon makes sense for long, hard tasks where a better answer is worth more tokens: reviewing a large codebase, deep research across many documents, or security work. For a quote follow-up email or a product description, a lighter model is usually enough.
If you run a professional services firm in Melbourne or a wholesaler in Brisbane and your systems use Gemini, ask your developer or software provider:
- Which Gemini model does each of our tools call, and is anyone planning to switch to Gemini 4 Carbon?
- What happens to our monthly AI cost when the introductory price ends on 30 November 2026?
- Have we measured the cost per task, not only the price per token, on our real work?
- Can any of our work move to the Batch rate, which Google lists at half the Standard price for jobs that do not need an immediate answer?
- Are there spending limits and alerts on the Google Cloud billing account, in case an agent runs longer than expected?
- If a tool reads emails, documents or web pages, what stops hidden instructions in that content from steering the model?
Developers should read Google's notes before switching. Google says an upgrade from Argon needs no changes to the request format, but the app must allow for longer thinking. Systems moving from Gemini 3 or earlier have more to change: explicit caching is not supported on Carbon, temperature, top_p and top_k settings are ignored, and Google recommends at least 32,768 for max_output_tokens and HTTP timeouts of at least 15 minutes. A system tuned for quick answers can time out if someone only swaps the model name. If you have no developer who can run these tests, a recruitment agency such as Talynx can help you hire one.
A paid API key that can call a powerful model is worth stealing, so the ASD's advice on protecting AI accounts and API keys applies directly. And if you are reviewing AI spend, our look at pay-per-use AI pricing compares token billing with per-user plans, and our post on OpenAI's GPT-4o API retirement shows why every model your software depends on needs a plan for change.
Comingwave builds and looks after business systems and integrations and custom software for SMEs, including tools that call AI models, and we can check how your systems use them and what they cost. Ask us for a free first consultation and a written quote.
Frequently asked questions
What is Gemini 4 Carbon?
Gemini 4 Carbon is Google's strongest AI model, built for cyber security work and long, multi-step agent tasks in knowledge work, software engineering and science. It thinks for longer than Gemini 4 Argon on each request.
Who can use Gemini 4 Carbon now?
Google says all paying developers and enterprises can use it through the Gemini API in Google AI Studio and Gemini for business, and it is rolling out to Google AI Ultra subscribers. It is not on the Gemini API's Free tier.
How much does Gemini 4 Carbon cost?
Until 30 November 2026 it costs $2 per million input tokens and $10 per million output tokens. From 1 December 2026, Google's pricing page lists $4 and $20, and thinking tokens are billed as output.
Should my business switch from Gemini 4 Argon to Carbon?
Not by default. Google says Argon remains the ideal choice for most software engineering and knowledge work. Test Carbon on your hardest tasks, compare the cost per task and the quality, then decide.
What is /swarm in Google Antigravity?
It is a preview feature in Google Antigravity, used by developers as an app, a command-line tool and editor extensions, that turns one prompt into a coordinated team of AI agents. Google's own team warns that this uses quota much faster than a normal chat.
Key takeaways
- Gemini 4 Carbon opened to all paying developers and enterprises on 10 November 2026.
- The introductory price of $2 and $10 per million tokens ends on 30 November 2026; from 1 December 2026 it is $4 and $20.
- Carbon uses more tokens per task than Argon, so measure cost per task, not just price per token.
- Google still recommends Argon for most everyday work.
- Ask your developer which model your tools use and set billing limits before the price change.