Gemini 4 Carbon: Google's Strongest Model, Announced Oct 28
Gemini 4 Carbon is Google's new strongest model, for cyber defense and long coding tasks. Who gets it first, API pricing, how it differs from Argon, what to do.
PromptWises is reader-supported. Some links may earn us a commission at no extra cost to you. This never affects our ratings or recommendations.
Update (11 November 2026): Carbon became generally available on November 10. It is now on the paid Gemini API and in Gemini for business, and rolling out to Google AI Ultra subscribers and GitHub Copilot paid plans; the introductory $2 / $10 price runs through November 30. See Gemini 4 Carbon now available. The rest of this post describes the October 28 announcement.
Google announced Gemini 4 Carbon on October 28, 2026 and calls it “our strongest model”, built for cybersecurity and long-horizon agentic coding. You almost certainly cannot use it yet. Access starts with a subset of cyber defenders, and Google gives no date for developers, businesses or consumers. When it does reach the API, it starts at Argon’s introductory price of $2 per million input tokens and $10 per million output tokens, then doubles. The source is Google’s announcement by Koray Kavukcuoglu of Google DeepMind.
What Gemini 4 Carbon is
Carbon builds on Gemini 4 Argon, which became generally available two weeks earlier. Google pitches it for sustained engineering work, such as finding the root cause of a bug in a large codebase and patching it. Google says it has seen Carbon “run for days to persist through dead ends” on hard multi-step tasks.
Carbon uses the same layered safety stack as Argon: model training, real-time misuse defences, prompt-injection resilience and chain-of-thought monitoring. Google’s announcement does not give a model ID, context window or output limit, so we do not know yet whether those match Argon’s.
Who can use it now
- This week: a subset of cyber defenders.
- Next: more trusted partners, in phases, through the Fairwind Program. Its page gives priority to governments, critical infrastructure operators and core technology platforms, and applicants are vetted. On October 29 the page still named Argon as its main model.
- Later: developers, businesses and consumers “as soon as possible”, after more work on guardrails. No date.
Google also says it is taking part in the US government’s voluntary pre-release model testing process. Carbon is not in the Gemini app, and on October 29 neither the Gemini API changelog nor the pricing page listed it.
API pricing
Google has published the price ahead of general access. Per million tokens:
| Gemini 4 Carbon | Gemini 4 Argon (paid tier, at the time of writing) | |
|---|---|---|
| Input, introductory | $2.00 | $2.00 through Nov 30, 2026 |
| Output, introductory | $10.00 | $10.00 through Nov 30, 2026 |
| Cached input | 95% off input price | $0.10 through Nov 30 (implicit caching) |
| Input / output after intro | $4.00 / $20.00 | $4.00 / $20.00 from Dec 1, 2026 |
| Available | Cyber defenders only at announcement; paid API since Nov 10 | Paid API; Google AI Pro and Ultra in the app |
At announcement, the Carbon footnote said only that the $4 and $20 prices apply “after the introductory period expires”. Since general availability, the pricing page sets the same November 30 cut-off as Argon. Google has not published Batch, Flex or Priority rates for Carbon.
Google’s benchmark claims
Google reports that Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59. On security tests it reports 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym and 71.0% on CWE-bench (patching). On WebDev Arena it gained 89 Elo to 1768, fourth place. Google says the external cybersecurity benchmarks were evaluated without guardrails.
On safety, Google reports a 0.0% attack success rate on Gray Swan’s indirect prompt-injection benchmark (Argon: 0.7%) and, in internal tests, about half as many unsupported claims as Argon. All of these are Google’s numbers on tests Google chose. Our guide to reading AI model announcements explains why you should check them against your own work.
What to do now
- Most readers: since November 10, Carbon is on the paid API and rolling out to AI Ultra. On AI Pro, Gemini 4 Argon remains the model you get.
- Security teams at eligible organisations: read the Fairwind Program page and apply through its interest form.
- Developers budgeting for 2026: plan on $2 / $10 at launch and $4 / $20 later, and keep Argon’s December 1 price rise in mind.
- Developers who want a cheaper option: Google’s mid-tier Gemini 4 Neon, released on November 18, costs $3 / $15 per million tokens and is worth testing before Carbon’s price rises on December 1.
For choosing between assistants today, see our Gemini review and the ChatGPT vs Claude vs Gemini comparison. The general availability details are in our follow-up post. More launches are in our news section.
Frequently asked questions
Can I use Gemini 4 Carbon now?
Yes, since November 10, 2026, if you pay for the Gemini API or Gemini for business. It is rolling out to Google AI Ultra subscribers in the Gemini app, AI Mode and Antigravity, and to GitHub Copilot paid plans. See our post on Carbon's general availability for the details.
How much will the Gemini 4 Carbon API cost?
Google says Carbon launches at the same introductory price as Argon: $2 per 1M input tokens and $10 per 1M output tokens, with cached input tokens 95% off the input price. After the introductory period it rises to $4 and $20. Google's pricing page now says the introductory price runs through November 30, 2026, and $4 / $20 applies from December 1.
What is the difference between Gemini 4 Carbon and Gemini 4 Argon?
Carbon builds on Argon, which became generally available on October 13, 2026, and Google calls it its strongest model, aimed at cybersecurity and long-horizon coding. It shares Argon's safety stack and introductory price. Google still recommends Argon for most work and Carbon for the hardest tasks, because Carbon spends more thinking tokens per turn.
When will Gemini 4 Carbon be available to everyone?
It became generally available on November 10, 2026: on the paid Gemini API and AI Studio, in Gemini for business, and rolling out to Google AI Ultra subscribers. It is not on the API free tier, and Google names only Ultra for the consumer apps.
Are the Gemini 4 Carbon benchmarks reliable?
They are Google's own reported results, and Google says the external cybersecurity benchmarks were run without guardrails, so the model you eventually get may behave differently. Treat them as a signal, not proof, and test on your own work once you have access.