Gemini 4 Carbon: what it is, how it scores, and how to get access
Gemini 4 Carbon scores 59 on the Artificial Analysis Intelligence Index, #1 at launch. There is no public API yet — here is how access works, what it will cost, and what to use today.
29 oct 2026 · API ManyGoogle announced Gemini 4 Carbon on October 28, 2026, two weeks after Gemini 4 Argon. Google describes it as its strongest model, with state-of-the-art cybersecurity and long-horizon capabilities, built for sustained agentic work such as root-causing and patching bugs across large codebases.
Source:Google: Gemini 4 Carbon announcement (Oct 28, 2026)
Gemini 4 Carbon at a glance
| Gemini 4 Carbon | Source | |
|---|---|---|
| Announced | October 28, 2026 | |
| Artificial Analysis Intelligence Index v5 | 59 — #1 at launch | Artificial Analysis |
| Context window | 1M tokens | Artificial Analysis |
| Input / output | Text, image, video and speech in; text out | Artificial Analysis |
| Introductory price | $2 / $10 per 1M input / output tokens | |
| Price after the introductory period | $4 / $20 per 1M input / output tokens | |
| Cached input | 95% off the input price | |
| Public API | No — not yet publicly available | Artificial Analysis |
| Published model ID | None yet | Gemini API models list |
How access works
- Now: Google says it is making Carbon available to a subset of cyber defenders, starting the week of October 28.
- Next: access is phased to additional trusted partners through the Fairwind Program, which gives high-priority defenders such as governments, healthcare providers and telecommunications services early access to advanced models. Organizations apply; access is managed and cannot be shared or resold.
- Later: Google says it will bring Carbon to developers, businesses and consumers via its product surfaces “as soon as possible” and “soon”. No date has been given.
- Model ID: Google has not published one. As of October 29, 2026 the Gemini API model list and the Google DeepMind model-card index do not include Carbon, and Artificial Analysis marks it “No publicly available API”.
Sources:Google: Gemini 4 Carbon announcement (Oct 28, 2026) · Google DeepMind: Fairwind Program · Google AI for Developers: Gemini API models · Google DeepMind: model cards · Artificial Analysis Intelligence Index leaderboard
What it will cost
Google says Carbon launches at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input 95% off. A footnote in the announcement says $4 / $20 applies after the introductory period. Artificial Analysis describes the same offer as a 50% launch discount “with no confirmed end date” and uses the $4 / $20 standard price in its charts. Google has not said when the introductory period ends, so plan on $4 / $20.
Sources:Google: Gemini 4 Carbon announcement (Oct 28, 2026) · Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026)
Independent benchmarks (Artificial Analysis)
- Intelligence Index v5: 59, the #1 score at launch, with the top score in 4 of the 9 Index evaluations — HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA.
- AA-Omniscience v1.1 (knowledge): 54, with 58% accuracy and an 11% hallucination rate, the second lowest of any model.
- AA-Briefcase (agentic knowledge work): 1609 Elo, 126 points above Gemini 4 Argon but not the leader.
- Cost to run the Index at standard pricing: $7.76 per task, using 123k output tokens per task.
Sources:Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026) · Artificial Analysis Intelligence Index leaderboard
For how these numbers stack up against other models, see the comparisons below: one ranks Carbon against Claude Opus 5.5 and GPT-6 Astra on overall score, cost and hallucination; the other goes deep on Carbon versus Opus 5.5 for coding agents and security work.
Results reported by Google
- Cybersecurity, evaluated without guardrails: 88.2% on SEC-bench Pro (finding vulnerabilities), 54.8% on ExploitGym (exploitation) and 71.0% on Collinear’s CWE-bench (patching).
- Web development: +89 Elo, to #4 (1768) on WebDev Arena.
- Shortcut-taking: successful cheating on Terminal-Bench 4.0 cut to 0.0%, with cheating attempts at 0.4%.
- Prompt injection: 0.0% attack success on Gray Swan’s Indirect Prompt Injection benchmark, versus 0.7% for Argon.
Source:Google: Gemini 4 Carbon announcement (Oct 28, 2026)
These are Google’s own measurements and have not been independently reproduced by Artificial Analysis.
What you can use today
While Carbon is gated, these models are live on API Many through one OpenAI-compatible endpoint: Gemini 4 Argon, the previous Gemini frontier model; Claude Opus 5.5; and GPT-6 Astra.
Related
- Gemini 4 Carbon vs Claude Opus 5.5 vs GPT-6 AstraOverall index, cost per task, hallucination and which to pick by use case.
- Gemini 4 Carbon vs Claude Opus 5.5Coding agents, long tasks, security and availability — plus what changed from Argon.
FAQ
Is there a Gemini 4 Carbon API?
Not a public one. Google is rolling Carbon out to a subset of cyber defenders and then to trusted partners through the Fairwind Program. Developers, businesses and consumers are promised access soon, with no date given.
What is the Gemini 4 Carbon model ID?
Google has not published one. Treat any “gemini-4-carbon” ID offered today as unofficial.
How much will Gemini 4 Carbon cost?
Google lists an introductory $2 / $10 per million input / output tokens, then $4 / $20. Cached input is 95% off. The end of the introductory period has not been announced.
Can I use Gemini 4 Carbon on API Many?
No. Carbon is not available on API Many. Gemini 4 Argon, Claude Opus 5.5 and GPT-6 Astra are.
How is Carbon different from Gemini 4 Argon?
Artificial Analysis scores Carbon 59 on its Intelligence Index versus 50 for Argon, at the same $4 / $20 standard price. Argon is available now; Carbon is not.
Sources
- Google: Gemini 4 Carbon announcement (Oct 28, 2026) — Announcement date, access plan, prices, cybersecurity, WebDev Arena and safety results.
- Artificial Analysis: Gemini 4 Carbon is the new #1 (Oct 28, 2026) — Index score, per-evaluation leads, AA-Omniscience, AA-Briefcase, cost per task, context, modalities, launch discount.
- Artificial Analysis Intelligence Index leaderboard — Live leaderboard (59, #1) and the “No publicly available API” label.
- Google DeepMind: Fairwind Program — Who the Fairwind Program is for and how access is managed.
- Google AI for Developers: Gemini API models — Public Gemini API models; Carbon not listed as of October 29, 2026.
- Google DeepMind: model cards — No Carbon model card as of October 29, 2026.
Todos los modelos de este artículo están en API Many. Pago por uso desde $10.