Gemini 4 Carbon Is Number One on Intelligence Index v5 at 59. Access Goes to Fairwind Cyber Defenders First, and the Index Run Burned 200 Million Output Tokens.
Google announced Gemini 4 Carbon on October 28, two weeks after Argon. Artificial Analysis scores it 59 on Intelligence Index v5, one point above Claude Opus 5.5 at 58, leading four of nine evaluations. It is not the coding leader: 59 in Antigravity against 62 for Claude Code with Opus 5.5 on the Coding Agent Index. The Index run used 200 million output tokens against a 64 million median, $7.76 per task at the $4 and $20 list. Access is a subset of cyber defenders through the Fairwind Program, then trusted partners, with no date for everyone else. Intro pricing is $2 and $10 with no published end date. Not to be confused with Hugging Face's Carbon-X. We read Google's post and AA's article and model page.
By T.W. Ghost
Google announced Gemini 4 Carbon today, October 28, 2026, about two weeks after Gemini 4 Argon, and Artificial Analysis has already put it at the top of the board it rebuilt yesterday. On Intelligence Index v5, Carbon at high effort scores 59, one point above Claude Opus 5.5 at max, which was the leader for one day. It leads four of the nine evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, and AutomationBench-AA.
Two catches travel with the number. The first is access: Google is releasing Carbon to a subset of cyber defenders through its Fairwind Program first, then to trusted partners, and to everyone else "as soon as possible," with no date. The second is on AA's model page: Carbon generated 200 million output tokens to run the Index, against a median of 64 million, which puts its cost per Index task at $7.76 at the standard $4 and $20 list price. The one-point lead is real on AA's scale; most readers cannot call the model, and the ones who can will pay for its verbosity. This is Google's Carbon; Hugging Face's DNA model Carbon-X is unrelated. We read Google's post, AA's article, and AA's model page. Here is what held up.
What Google Shipped, and to Whom
| Gemini 4 Carbon | |
|---|---|
| Announced | October 28, 2026 |
| Access | Fairwind Program cyber defenders first; then trusted partners; "as soon as possible" for others, no date |
| Intro price | $2 in / $10 out per 1M tokens, then $4 / $20; cached input 95% off (Google) |
| Intro end date | None published (Argon's intro ends November 30; Carbon's is unstated) |
| AA Intelligence Index v5 | 59 (high), number one; Opus 5.5 (max) 58 |
| Leads | 4 of 9 v5 evaluations |
| AA Coding Agent Index | 59 (high, Antigravity CLI) vs Claude Code with Opus 5.5 (max) 62 |
| AA-Briefcase | 1,609 Elo (Argon: 1,483); behind Grok 4.7, Muse Wonder, Kimi K3.1 Pro on that board |
| Index run | 200M output tokens (median 64M); $7.76 per task; about 99.8 tokens per second |
| Google's cyber numbers | SEC-bench Pro 88.2%; ExploitGym 54.8%; CWE-bench 71.0% (vendor) |
Number one by a point, on a one-day-old scale. v5 shipped October 27. Carbon's 59 and Opus's 58 are both v5 numbers, and Opus is measured at max effort. The lead is within the kind of margin that a re-run moves.
Not the coding leader. On AA's agent board Carbon in Antigravity scores 59 and Claude Code with Opus 5.5 scores 62. Carbon's wins are on science, knowledge, hallucination, and automation evaluations, not the coding and knowledge-work stack.
The token count is the cost story. Two hundred million output tokens is more than three times the median run. At $4 and $20 that is $7.76 per Index task; at the intro price it would be less, and Google has not said when the intro price ends.
The Gate
Fairwind is the same shape as the gates we covered this month at the other labs: OpenAI's Critical cyber rating on Terra, Anthropic's three-tier Cyber Verification Program. Google's version puts defenders first and names no eligibility rules in the launch post. Carbon's vendor cyber scores, SEC-bench Pro at 88.2% and ExploitGym at 54.8%, are Google's measurements with no rival rows printed, and they are the reason the gate exists.
Verified vs Unconfirmed: The Scorecard
| Claim | Verdict |
|---|---|
| Announced Oct 28, 2026 | Verified (Google blog) |
| v5 Intelligence Index 59, number one; Opus 5.5 (max) 58; leads 4 of 9 | Verified (AA article and model page) |
| Coding Agent Index 59 vs 62 | Verified (AA article) |
| AA-Briefcase 1,609 Elo, Argon 1,483 | Verified (AA article) |
| 200M output tokens; 64M median; $7.76 per task; about 99.8 tokens per second | Verified (AA model page) |
| Intro $2 / $10 then $4 / $20; cached input 95% off | Verified (Google blog) |
| Fairwind first, trusted partners, "as soon as possible" | Verified (Google blog) |
| SEC-bench Pro 88.2%; ExploitGym 54.8%; CWE-bench 71.0% | Vendor (Google) |
| Intro end date | Unverified (none published) |
| Model card or safety report detail | Unverified (not read this pass) |
Who Should Use It
Use Carbon if you are inside Fairwind or a trusted-partner program and your work is the science, knowledge, or automation evaluations it leads; budget on 200 million tokens per Index-sized workload, not on the per-token price.
Skip it as a routing target until there is a public API, and skip the "number one" line without the scale and the margin: v5, one point, against a max-effort Opus.
The frontier is elsewhere, and for most readers it is on sale: Argon is the Gemini 4 you can call today. Our model comparison has the current standings side by side.
Sources
- Google: introducing Gemini 4 Carbon
- Artificial Analysis: Gemini 4 Carbon is the new number one
- Artificial Analysis: Gemini 4 Carbon model page
*Not sure whether a gated number-one model or a shipping number-four fits your work? Take the free 2-minute quiz and get matched. Then read about the index Carbon leads and what is private inside it: Artificial Analysis Intelligence Index v5.*