Google's Gemini 4 Carbon Takes No. 1 on Artificial Analysis Intelligence Index v5, With Access Limited to Cyber Defenders
Google's new frontier model edges Claude Opus 5.5 by one point at 59, posts strong cybersecurity scores, and will reach developers only after a phased rollout.
Published
Last updated / Fact-checked on
Google's Gemini 4 Carbon is the new No. 1 on the Artificial Analysis Intelligence Index v5, scoring 59. That puts it one point ahead of Anthropic's Claude Opus 5.5 (max), which took the top spot when the index launched a day earlier. Google announced Carbon on Oct. 28, two weeks after Gemini 4 Argon. For now, most people can't use it. Google is giving access first to a small group of cybersecurity defenders.
The new leaderboard
| Model | Intelligence Index v5 |
|---|---|
| Gemini 4 Carbon (high) | 59 |
| Claude Opus 5.5 (max) | 58 |
| GPT-6 Astra (max) | 57 |
| GPT-6.1 Sol (max) | 55 |
| Gemini 4 Argon (high) | 50 |
Source: Artificial Analysis. Scores other than Carbon's are from the Oct. 27 v5 launch.
According to Artificial Analysis, Carbon has the top score in four of the index's nine evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA. It doesn't lead in coding or knowledge work, though it improved a lot over earlier Gemini models in both. On Artificial Analysis's separate Coding Agent Index, Carbon scores 59, behind Claude Code with Opus 5.5 at 62.
A one-point lead is narrow. Carbon is the top scorer under this method, not automatically the best model for every job. For how v5 works, including why 60% of it now rests on private tests, see our explainer on Intelligence Index v5.
Cybersecurity is the headline, and the reason for the slow rollout
Google pitches Carbon as its strongest model for cybersecurity and long, multi-step work. It reports these scores on external benchmarks, which it notes were run without guardrails:
- SEC-bench Pro (finding vulnerabilities): 88.2%
- ExploitGym (confirming a flaw can be exploited): 54.8%
- CWE-bench (writing patches): 71.0%
Google also says its own teams have seen Carbon work for days on hard engineering problems, such as tracking down and fixing subtle bugs in large codebases. That claim comes from internal use, not independent testing.
Skills that help defenders find and confirm flaws could also help attackers. Google says it's taking a "deliberate approach" to the rollout and is taking part in the U.S. government's voluntary pre-release testing. Starting this week, a subset of cyber defenders gets access, followed gradually by more trusted partners through its Fairwind Program. Google hasn't set a date for developers, businesses or consumers. It says only "as soon as possible."
What it will cost
| Input (per 1M tokens) | Output (per 1M tokens) | |
|---|---|---|
| Introductory price | $2 | $10 |
| Standard price | $4 | $20 |
Cached input tokens are 95% off. Artificial Analysis says the introductory discount has no confirmed end date.
At standard prices, Artificial Analysis puts Carbon's cost at $7.76 per Intelligence Index task. That's about 23% less than Claude Opus 5.5 (max) at $10.04. It's about the same as Gemini 4 Argon ($7.68), which scores 9 points lower. Carbon also used about 29% fewer output tokens per task than Opus 5.5. These are costs on benchmark tasks, not a forecast for any real app.
What this means
For developers, the bigger story may be the work Carbon is built for: long, unsupervised engineering work, plus a security process that runs from finding a flaw to fixing it. If independent testing backs up those claims, that matters more than a one-point lead.
For now, there's little to act on. Until access widens, almost no one outside Google's first group can test Carbon on their own code. When it opens up, compare it with Opus 5.5 and GPT-6.1 Sol on your own tasks, at the effort level you would actually use, and track success rate and total cost.
How we reported this
We checked every score, price and access detail against Google's announcement and Artificial Analysis's Carbon write-up. We cross-checked Google's figures with 9to5Google's report. DailyMech has not tested Carbon, because it isn't publicly available yet.
Sources
- Google, "Gemini 4 Carbon: our next frontier model" (Oct. 28, 2026): https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/
- Artificial Analysis, "Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5" (Oct. 28, 2026): https://artificialanalysis.ai/articles/gemini-4-carbon-new-number-1
- Artificial Analysis, "Announcing Artificial Analysis Intelligence Index v5" (Oct. 27, 2026): https://artificialanalysis.ai/articles/artificial-analysis-intelligence-index-v5
- 9to5Google, "Google announces Gemini 4 Carbon with 'significant gains across coding and agentic tasks'" (Oct. 28, 2026): https://9to5google.com/2026/10/28/gemini-4-carbon-announcement/
DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.