· reAPI Team
Gemini 4 Carbon: Release Date, Price, Benchmarks and Access
Gemini 4 Carbon was announced on October 28, 2026. Who can use it now, its $2/$10 introductory price, Google's benchmark claims, independent scores and Argon.
Google announced Gemini 4 Carbon on October 28, 2026, calling it "our strongest model with state-of-the-art cybersecurity and long horizon capabilities". It is not generally available yet. Google is giving it first to "a subset of cyber defenders", then to more trusted partners through its Fairwind Program, and only after that to developers, businesses and consumers, with no date given.[1] When it opens, it will cost $2 per million input tokens and $10 per million output tokens at the introductory rate, the same as Gemini 4 Argon.[1]
Google says Carbon builds on Gemini 4 Argon, "which was launched two weeks ago".[1] This guide covers what Google has published about access, price and benchmarks, what an independent tester measured, and what to use while you wait, as checked on November 4, 2026.
When can you use Gemini 4 Carbon?
Google describes a staged rollout:[1]
- From launch week: a subset of cyber defenders.
- Next: additional trusted partners through the Fairwind Program.
- Later: developers, businesses and consumers "via our product surfaces as soon as possible", after more feedback and work on guardrails.
Google also says it is taking part in the U.S. government's voluntary process for pre-release model testing. On November 4, 2026, Carbon did not appear on Google's Gemini API pricing page, which listed Gemini 4 Argon as the only Gemini 4 model.[2] If you need an API today, Carbon is not yet an option.
Gemini 4 Carbon price
| Introductory price | After the introductory period | |
|---|---|---|
| Input, per 1M tokens | $2 | $4 |
| Output, per 1M tokens | $10 | $20 |
| Cached input | 95% off the input price | 95% off the input price |
Source: Google's announcement and its footnote.[1] Google has not said how long Carbon's introductory period will last. For comparison, Argon's introductory rate runs through November 30, 2026, after which it moves to the same $4 and $20.[2]
For a sense of scale, a request with 20,000 input tokens and 2,000 output tokens would cost 0.02 × $2 + 0.002 × $10 = $0.06 at the introductory rate and $0.12 at the standard rate. Long agent runs generate far more output than that, and Artificial Analysis measured Carbon as unusually verbose, so plan budgets on output, not input.[3]
Gemini 4 Carbon benchmarks
Google's announcement leads with security results, all "evaluated without guardrails":[1]
| Benchmark | What it tests | Carbon |
|---|---|---|
| SEC-bench Pro | Finding real vulnerabilities | 88.2% |
| ExploitGym | Validating vulnerabilities through exploitation | 54.8% |
| CWE-bench (Collinear) | Patching at the root cause | 71.0% |
| WebDev Arena | Web development, human-rated | #4, 1768 Elo (+89) |
Google adds that on internal benchmarks Carbon more than doubles Argon's success rate at generating secure code, that it reduced successful cheating on Terminal-Bench 4.0 to 0.0%, and that it had a 0.0% attack success rate on Gray Swan's indirect prompt-injection benchmark, down from Argon's 0.7%.[1] These are Google's own reported results.
Long-horizon work and behavior
Beyond scores, Google describes Carbon as built for "sustained agentic workflows". Its teams have seen it run for days on multi-step engineering tasks such as root-causing and patching subtle bugs across large codebases. Google also reports that Chrome engineers found it more than 10 times better than another lab's large frontier model at generating high-quality patches, and that Wiz sees it leading web penetration-testing benchmarks.[1]
The behavior claims are specific. Compared with Argon, Google says Carbon cuts unsupported claims by about half, reduces overreach by 26% and under-service by 29%, and in live coding with internal teams saw high-severity misalignment alerts drop by almost 80%. It is also trained to surface discrepancies to the user before acting instead of guessing.[1] All of these are Google's internal measurements.
Independent results
Google cited Artificial Analysis, which tests models on one shared index. At launch Google reported Carbon at #1 with a score of 59.[1] On November 4, 2026, Artificial Analysis listed Gemini 4 Carbon (High) first of 62 models with a score of 60.[3] Its page adds three numbers worth knowing:
- It generated 200 million output tokens to complete the index, which Artificial Analysis calls "very verbose" against a median of 73 million.
- Output speed was 88.7 tokens per second.
- The cost to run each index task was $7.76, calculated at the $4 and $20 standard rates.
Artificial Analysis also lists text, image, speech and video input, text output and a 1M-token context window.[3]
High verbosity matters for budgets: output is the expensive side of the bill, so a model that writes more per task can cost more per task than its token price suggests.
Gemini 4 Carbon vs Argon
The two share a price and a context class, but not availability. Argon is open to developers now through the Gemini API and Google AI Studio, while Carbon is limited to the early cohorts above.[1][2] Carbon is the stronger model on Google's security benchmarks and on Artificial Analysis's index. Our three-way benchmark comparison covers Argon against GPT-6 Astra and Claude Opus 5.5, and our Gemini 4 Argon pricing guide explains its billing.
FAQ
Is Gemini 4 Carbon available in the API?
No, not as of November 4, 2026. Google is limiting access to selected cyber defenders and Fairwind partners first, and it was not listed on the Gemini API pricing page.[1][2]
What is the Gemini 4 Carbon release date?
Google announced it on October 28, 2026, with early access for a subset of cyber defenders starting that week. It has not given a date for developers or consumers.[1]
What inputs does Gemini 4 Carbon accept?
Artificial Analysis lists text, image, speech and video input with text output and a 1M-token context window. Google's announcement itself does not state a context window.[3][1]
Is Gemini 4 Carbon better than GPT-6 Astra?
On Artificial Analysis's index on November 4, 2026, Carbon (High) scored 60 against 57 for GPT-6 Astra (Max).[3] On your own tasks, test both once Carbon is available.
Using Gemini 4 Argon while Carbon rolls out
Until Gemini 4 Carbon opens to developers, the closest model you can call is Gemini 4 Argon, at the same introductory price. reAPI serves Gemini 4 Argon at $0.80 per million input tokens and $4.00 per million output tokens on its model page, checked November 4, 2026.
References
- Google. Gemini 4 Carbon: our next frontier model. October 28, 2026. Retrieved November 4, 2026 from blog.google
- Google. Gemini Developer API pricing. Retrieved November 4, 2026 from ai.google.dev/gemini-api/docs/pricing
- Artificial Analysis. Gemini 4 Carbon (High) model page. Retrieved November 4, 2026 from artificialanalysis.ai/models/gemini-4-carbon