Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5
On October 28, 2026, Google DeepMind announced Gemini 4 Carbon, which it describes as its strongest model so far. Artificial Analysis, an independent evaluator, gives Carbon 59 points on its Intelligence Index v5, one point above Anthropic's Claude Opus 5.5 (58) and two above OpenAI's GPT-6 Astra (57). Gemini 4 Argon, which reached paying users about two weeks earlier, scored 50 on the same version. Carbon ranks first in four of the index's nine evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA.
Artificial Analysis puts the cost at $7.76 per index task at standard prices, against $10.04 for Claude Opus 5.5 and $6.25 for GPT-6 Astra. According to R&D World, Carbon produced about 200 million output tokens, roughly three times the median for comparable models, and took a median 97.5 seconds to deliver its first output token on Google's API, versus 3.6 seconds for peers. In the Coding Agent Index, Carbon scores 59 with Antigravity CLI, while Claude Code running Opus 5.5 scores 62.
On the updated AA-Omniscience v1.1 test, Carbon scores 54 against 48 for Opus 5.5 at max effort, and its accuracy rose to 58 percent from Argon's 48 percent. Its hallucination rate, however, climbed to 11 percent from Argon's 4 percent. Artificial Analysis measures that rate among answers that were not fully correct, so it is not the share of all questions answered wrongly. Access starts with cybersecurity defenders in Google's Fairwind program, and Google has not said when others will get it.
Why it matters
For teams building on frontier models, a one-point lead on a composite index is a weak basis for choosing a model. The more useful signal is the trade-off: Carbon leads on knowledge and reasoning tests, but its higher hallucination rate matters for any product that answers from its own knowledge, such as support bots or research tools. Its per-task cost sits below Opus 5.5, yet its token volume and first-token latency are heavy, so cost and speed should be tested against your own workload before any switch.
For founders and investors, the release shows how fast the top position changes hands, since Argon was the flagship only two weeks earlier. Google says API pricing will match Argon's introductory price, but it has not said when API access arrives, so roadmaps that depend on Carbon need a fallback. Companies selling security tooling should watch closely, because Google is leading with cybersecurity results that the index does not measure.
Key points
- 1Carbon scores 59 on Artificial Analysis Intelligence Index v5, versus 58 for Claude Opus 5.5 and 57 for GPT-6 Astra.
- 2Carbon ranks first in four of nine evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience and AutomationBench-AA.
- 3Artificial Analysis puts Carbon's cost at $7.76 per index task at standard prices, versus $10.04 for Opus 5.5.
- 4The AA-Omniscience hallucination rate rose to 11 percent from Argon's 4 percent, though it remains the second lowest listed.
