Google announced Gemini 4 Carbon on October 28, describing it as its strongest model with state-of-the-art cybersecurity and long horizon capabilities (Google). The company reported that Carbon ranks #1 on v5.0 of the Artificial Analysis Intelligence Index at 59 (Google). Tom’s Hardware reported that Artificial Analysis lists Carbon as the highest-ranked model on that index, one point ahead of Claude Opus 5.5 (Tom’s Hardware). A separate benchmark from SemiAnalysis ranks Claude Opus 5.5 first, with Carbon close behind (9to5Google).
Key Facts
| Item | Detail | Source |
|---|---|---|
| Announced | October 28 by Google | |
| Index score | 59 on Artificial Analysis Intelligence Index v5.0, ranked #1 | |
| Access | A subset of cyber defenders first, then trusted partners through the Fairwind Program | |
| Introductory price | $2 per million input tokens and $10 per million output tokens | |
| GDPx cost per task | Just over $16 for Carbon, around $51 for Claude | 9to5Google |
Timeline
- 2026-10-28: Google announced Gemini 4 Carbon (Google).
- 2026-11-02: 9to5Google reported Carbon’s results on the SemiAnalysis GDPx benchmark (9to5Google).
- 2026-11-03: Tom’s Hardware reported Carbon’s top position on the Artificial Analysis Intelligence Index V5 (Tom’s Hardware).
Details (as Reported)
Google reported that Carbon scored 88.2% on SEC-bench Pro, 54.8% on ExploitGym and 71.0% on CWE-bench, all evaluated without guardrails (Google). The company also reported a gain of 89 elo points to #4 (1768) on WebDev Arena (Google). In its internal testing, Google reported that Carbon cuts unsupported claims by approximately 2x over Gemini 4 Argon and reached a 0.0% attack success rate on the Gray Swan Indirect Prompt Injection benchmark (Google).
Tom’s Hardware reported that Carbon takes the number one score on four of the nine evaluations in the index (Tom’s Hardware). It also reported a score of 54 on the AA-Omniscience Index, which tracks hallucinations and recall, on a scale of -100 to +100 (Tom’s Hardware). Artificial Analysis found Carbon is not as good as Claude Code with Claude Opus 5.5 at programming, and that it falls behind Grok 4.7, Muse Wonder and Kimi K3.1 Pro on knowledge work, according to Tom’s Hardware (Tom’s Hardware).
The SemiAnalysis GDPx benchmark includes 206 tasks across finance, materials science, cybersecurity, software engineering, legal, biology and other domains (9to5Google). GPT-6 Astra placed third at about $54 per task (9to5Google). Removing the cybersecurity, biology and alignment domains puts GPT-6 Astra on top, with GPT 6.1 Sol and Claude Sonnet 5.5 ranking above Carbon (9to5Google). On cybersecurity alone, the benchmark found Meta’s Muse models best by a meaningful margin, with both Google models ahead of OpenAI and Anthropic models (9to5Google).
What Is Not Yet Confirmed
- The public release date has not been confirmed, and Carbon is coming soon for most customers (9to5Google).
- Google plans to gather feedback and iterate on guardrails before making Carbon available to developers, businesses and consumers as soon as possible (Google).
- Google is engaged in the U.S. government’s voluntary process for pre-release model testing (Google).