Gemini 4 Carbon is now generally available to paying developers in the Gemini API. Google’s release notes list it as GA on November 10, 2026, at an introductory price of $2 per million input tokens and $10 per million output tokens through November 30. On December 1, the price doubles to $4 and $20. In the Gemini app, Carbon is limited to Google AI Ultra subscribers, and a separate cyber version stays behind Google’s Fairwind gate.
Google first announced Carbon on October 28 as a gated model for cyber defenders. Two weeks later it opened the model to developers, businesses and consumers. We covered the earlier Gemini 4 models in what Gemini 4 Argon is, Argon’s general availability and Gemini 4 Hydrogen. This post covers where Carbon is available, what it costs before and after December 1, and how it compares with Argon and Hydrogen. Facts are as of November 16, 2026.
The short version: Carbon is Google’s strongest model, and it is built to spend more thinking tokens than Argon at the same per-token price. That makes it the most expensive Gemini 4 model per task. There is no free tier. Use Hydrogen or Argon for everyday work and save Carbon for the hardest debugging, security and research jobs. If you test Carbon this month, budget for the December 1 price, not the current one.
What Google released with Gemini 4 Carbon
The Gemini API release notes entry for November 10 calls gemini-4-carbon "our most capable reasoning model." It lists the introductory price of $2 and $10 per million tokens through November 30, 2026.
Google’s November 10 post by Koray Kavukcuoglu says Google released Carbon to cyber defenders two weeks earlier through Fairwind. It says Argon "remains the ideal choice for the majority of software engineering and knowledge work." Carbon is for the hardest multi-step coding, debugging and research tasks.
The technical details come from Google’s What’s new in Gemini 4 Carbon guide:
- Context: one 1,048,576-token window shared by input, thinking and output.
- Long output: standard responses stop at 65,536 tokens. Background execution can run up to 1M tokens minus your input.
- Thinking levels: low, medium (the default) and high.
- Caching: implicit caching only, on prompts of at least 4,096 tokens, with a 95% discount on cache hits. Explicit caching is not supported.
- Sampling: temperature, top_p and top_k are ignored.
- Agents: the Antigravity agent in Gemini Managed Agents now supports Carbon.
Where Gemini 4 Carbon is available now
Access depends on where you use it. The table sums up each surface as of November 16, 2026.
| Where | Who can use Carbon | Google source |
|---|---|---|
| Gemini API and AI Studio | Paid tier only; not on the free tier | Pricing page and What’s new guide |
| Google Cloud (Gemini Enterprise Agent Platform) | Verified enterprise customers in select regions; GA since November 10 | Cloud model page |
| Gemini app | Google AI Ultra subscribers only | Gemini Apps Help |
| AI Mode in Search and Google Antigravity | Rolling out to Google AI Ultra subscribers | November 10 blog post |
| Carbon for cyber | Allowlist only, through the Fairwind Program | Cloud model page |
On Google Cloud, the Gemini 4 Carbon model page lists the global, us and eu regions. The old Vertex AI models address now redirects to this platform’s docs. The page says Carbon "is available to verified enterprise customers in select regions." It lists a knowledge cutoff of March 2026.
For the app, the About Gemini models help article shows Carbon only under Google AI Ultra. Google AI Pro gets Argon and Hydrogen, and Google AI Plus gets Hydrogen. The article says work and school accounts depend on the Workspace license.
The remaining gate: Carbon for cyber
Carbon itself is open. Its cyber version is not. The Cloud model page describes Gemini 4 Carbon for cyber (gemini-4-carbon-cyber-permissive) as a version "for defensive cybersecurity use cases." It is available behind an allowlist. To request it, contact your Google team or apply through the Fairwind Program. Access to earlier Gemini cyber models does not carry over.
Google’s November 10 post ties Carbon’s cyber benchmark leads to access through Fairwind. Its October 28 post gave the numbers: 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench, "evaluated without guardrails."
The Fairwind Program page now says Google works with "thousands of partners globally." It prioritizes governments, critical infrastructure operators and core technology platforms. Partners must use phishing-resistant MFA and limit access to internal security, incident response or penetration testing teams. You apply through your Google Cloud account team. If you do not qualify, the page points to CodeMender with publicly available models.
What Gemini 4 Carbon costs, before and after December 1
These prices come from the Gemini API pricing page, read November 16, 2026.
| Price per 1M tokens | Through Nov 30, 2026 | From Dec 1, 2026 |
|---|---|---|
| Input | $2.00 | $4.00 |
| Cached input (implicit caching) | $0.10 | $0.20 |
| Output, including thinking | $10.00 | $20.00 |
Google’s guide says every token is billed, including thinking tokens. Grounding with Google Search is billed separately: 5,000 free requests a month across Gemini 3 and newer models, then $14 per 1,000. The pricing page says paid-tier content is not used to improve Google’s products.
A worked cost example for a small-site task
Say an agency asks Carbon to review a plugin update for a client’s WordPress site before it ships. Each review sends 60,000 input tokens: the changed files, related code and instructions. It returns 12,000 output tokens, thinking included. These are our assumptions, not Google’s.
| Model and price period | Cost per review | 100 reviews a month |
|---|---|---|
| Gemini 4 Carbon, through Nov 30 | $0.24 | $24 |
| Gemini 4 Carbon, from Dec 1 | $0.48 | $48 |
| Gemini 4 Carbon, from Dec 1, 40,000 input tokens cached | $0.33 | $33 |
| Gemini 4 Hydrogen, same token counts | $0.18 | $18 |
The math from December 1: 60,000 input tokens at $4 per million is $0.24. 12,000 output tokens at $20 per million is another $0.24. If 40,000 of those input tokens repeat and hit the cache, they cost $0.008 instead of $0.16.
Two cautions. First, Argon has the same per-token price, but Google says Argon uses fewer thinking tokens per turn. The same review on Argon should cost less, so test both. Second, Carbon’s output count will swing with the thinking level you choose. Our guide to prompt caching covers how to keep repeated context cacheable. Our post on batch API discounts covers jobs that can wait.
Gemini 4 Carbon vs Argon and Hydrogen
Google’s pricing page and guides describe the three Gemini 4 models this way, as of November 16, 2026.
| Model | Price per 1M tokens (input / output) | Default thinking and free tier | Google’s suggested use |
|---|---|---|---|
| Gemini 4 Carbon | $2 / $10 through Nov 30; $4 / $20 from Dec 1 | Medium; no free tier | Root-cause debugging, architecture, dense documents, proofs, security audits |
| Gemini 4 Argon | $2 / $10 through Nov 30; $4 / $20 from Dec 1 | Medium; no free tier | Most advanced coding and knowledge work, with faster responses |
| Gemini 4 Hydrogen | $1.50 / $7.50 | High, with an added xhigh level; free tier | Everyday coding and agent workflows |
Hydrogen went GA on November 12, per the release notes. On cost, Hydrogen is the cheapest of the three and the only one with a free tier. Carbon and Argon share a price, so the difference is how many tokens each spends. Google’s guide says moving from Argon to Carbon "requires no request schema changes." You only change the model ID.
Google also reports new results for Carbon. Its November 10 post says Carbon passed 92% of a 375-task agentic honesty benchmark, against 70% for Argon. It repeats Carbon’s 0.0% attack success rate on Gray Swan’s prompt injection benchmark. It also says prompt injection robustness "remains an area for continued improvement." These are Google’s own numbers.
Gated cyber models at OpenAI and Anthropic
With Carbon now public, Google follows the same split as the other two labs: a public general model plus a gated cyber version.
- OpenAI. Its GPT-6 Cyber announcement says the model "is initially available through the API and Codex to a select group of approved Daybreak Red customers." OpenAI reports 48.5% on ExploitGym. Google reported 54.8% for Carbon on the same benchmark. Both are vendor-reported, and the setups may differ. See our post on GPT-6 Cyber.
- Anthropic. Its Fable and Mythos 5.2 announcement says Claude Mythos 5.2 "is available to vetted organizations that work in these dual-use areas." Fable 5.2 is available to all users. See Claude Mythos 5.2 and our GPT-6 Astra vs Claude Fable 5.2 comparison.
What to do now
- Do not make Carbon your default. Run your evaluations on Hydrogen and Argon first. Our Gemini model guide still explains the older lineup.
- Budget at the December 1 price. Price any Carbon work at $4 and $20 per million tokens. Set spend caps before you open it to a team.
- Fix your request settings. Google recommends a max_output_tokens of at least 32,768 and HTTP timeouts of at least 15 minutes. Remove explicit caching calls.
- Tell staff what they will see. Only Google AI Ultra subscribers get Carbon in the Gemini app. Pro users see Argon and Hydrogen.
- Ask about Fairwind only if it fits. A state agency, hospital or utility may qualify for Carbon for cyber. Most nonprofits and agencies will not.
Frequently Asked Questions
What is Gemini 4 Carbon?
It is Google’s strongest Gemini 4 model, built for long, hard tasks in coding, research, document work and cyber defense. Google announced it on October 28, 2026, and made it generally available on November 10.
Can I use Gemini 4 Carbon in the Gemini API now?
Yes, on the paid tier. The model ID is gemini-4-carbon. It is not available on the free tier.
How much does Gemini 4 Carbon cost?
Through November 30, 2026, it costs $2 per million input tokens and $10 per million output tokens. From December 1, it costs $4 and $20. Cached input is $0.10, then $0.20.
Is Gemini 4 Carbon in the Gemini app?
Only for Google AI Ultra subscribers, per Google’s help center. Google AI Pro subscribers get Argon and Hydrogen instead.
Is Gemini 4 Carbon available on Google Cloud?
Yes. Google’s Cloud model page lists it as GA since November 10 for verified enterprise customers in select regions.
Is anything about Carbon still gated?
Yes. Gemini 4 Carbon for cyber is allowlist only. You request it through your Google team or the Fairwind Program.
Is Carbon more expensive than Argon?
The per-token price is the same. Google says Carbon spends more thinking tokens per turn, so the same task usually costs more on Carbon.
How does Carbon compare with Gemini 4 Hydrogen?
Hydrogen costs $1.50 and $7.50 per million tokens and has a free tier. Google positions it for everyday coding and agents. Carbon is for the hardest tasks.
Should our nonprofit switch to Carbon?
Probably not for routine work. Test Hydrogen or Argon first and use Carbon only where they fall short.