Gemini 4 Carbon Tops Artificial Analysis Benchmarks

Google’s Gemini 4 Carbon has reached the top of the Artificial Analysis Intelligence Index v5 with a score of 59, driven by strong performance in complex reasoning, knowledge, and scientific tasks. The model ranks second only to Claude Opus 5.5 on the Coding Agent Index—a significant milestone given Google’s historical weakness in coding benchmarks. Currently rolling out to selected users with a 50% launch discount, Carbon leads 4 of 9 benchmark evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, and AutomationBench-AA.

On knowledge work, Carbon scores 1609 Elo on AA-Briefcase, surpassing its predecessor Gemini 4 Argon at 1483 Elo, though trailing Grok 4.7 (1638), Muse Wonder (1659), and Kimi K3.1 Pro (1678). The model demonstrates particular gains in presentation metrics and analytical quality. On the AA-Omniscience v1.1 benchmark, Carbon achieves 54 points with an 11% hallucination rate—second-lowest after Argon’s 4%—while reaching 58% accuracy.

Carbon costs $7.76 per Intelligence Index task, roughly 23% less than Claude Opus 5.5, while using 29% fewer output tokens. The model supports a 1M token context window with multimodal inputs (text, image, video, speech), pricing at $4/$20 per million input/output tokens under standard rates, and cached inputs receive a 95% discount.