Google’s Gemini 4 Carbon now tops Artificial Analysis’ Intelligence Index with a score of 59, a single point ahead of Claude Opus 5.5. Google DeepMind launched the model on 28 October, as reported by Tom’s Hardware.
Google is rolling Carbon out only to selected users, starting with vetted cyber defenders in its Fairwind Program, while Opus 5.5 has been on sale since 22 September. Developers can use Opus 5.5 today, but they can’t use Carbon yet.
Carbon also leads Artificial Analysis’ AA-Omniscience score, which rewards correct answers and penalises hallucinations. It scores 54 there, against 48 for Opus 5.5 and 46 for Gemini 4 Argon. Its hallucination rate is 11%, second lowest of any model behind Argon (4%) and well below Opus 5.5 and OpenAI’s GPT-6 Astra (both 50%).
| Model | Intelligence Index v5 | AA-Omniscience | Hallucination rate | Cost per index task |
|---|---|---|---|---|
| Gemini 4 Carbon (high) | 59 | 54 | 11% | $7.76 |
| Claude Opus 5.5 (max) | 58 | 48 | 50% | $10.04 |
| GPT-6 Astra (max) | 57 | – | 50% | $6.25 |
| Gemini 4 Argon | 50 | 46 | 4% | $7.68 |
Carbon answers more questions correctly than Argon (accuracy rose 10 points to 58%), but its hallucination rate also climbed from 4% to 11%. Overall, it takes first place in four of the index’s nine tests: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, and AutomationBench-AA.
With Google’s Antigravity CLI, Carbon scores 59 on the AA Coding Agent Index, three points behind Claude Code with Opus 5.5 (62). It’s second on Terminal-Bench 4.0 and FrontierCode, with Opus 5.5 leading both. Artificial Analysis notes that “Google has historically underperformed in this area and so this improvement stands out.”
On knowledge work, Carbon’s AA-Briefcase rating of 1609 is up 126 from Argon, but it sits behind xAI’s Grok 4.7 (1638), Meta’s Muse Wonder (1659), and Kimi K3.1 Pro (1678).
Against its closest rival, Carbon is also cheaper to run. It costs $7.76 per index task at standard pricing, about 23% less than Opus 5.5, and uses 123,000 output tokens per task, 29% fewer than Opus. Google charges $4 per million input tokens and $20 per million output tokens, the same as Argon, with a 50% launch discount bringing that to $2 and $10. Carbon has a 1M-token context window and accepts text, images, video, and speech.
Google is pitching Carbon mainly at cybersecurity and long-running tasks. Its own figures put the model at 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench, but those are Google’s internal measurements. Artificial Analysis ran the benchmarks, and no one else has repeated them.
Carbon arrives almost four weeks after Gemini 4 Argon, revealed on 1 October with the same cyber-first Fairwind rollout. In September, Anthropic launched Opus 5.5 with lower prices and tighter safeguards. It still leads on coding agents.
Google hasn’t given a date for general access, and Artificial Analysis says the 50% launch discount has no confirmed end date.
