Alphabet's Gemini 4 Carbon Tops AI Intelligence Index, Beating Claude and GPT Models

Alphabet (GOOGL) released its flagship Gemini 4 Carbon AI model on Wednesday, reclaiming the top spot on Artificial Analysis' Composite Intelligence Index with a score of 59, ahead of Claude Opus 5.5 at 58 and GPT-6 Astra at 57. Carbon, which follows the Gemini 4 Argon release by two weeks, scored 59 on the index versus Argon's 50. Google said the model reduces unsupported claims by about half and cuts unrequested overreach by 26%. It scored 88.2% on the SEC-bench Pro vulnerability-detection test and supports a 1 million token context window. Results are mixed across specific benchmarks: GPT-6 Astra leads FrontierSWE 2 software engineering at 65.5% versus Carbon's 61.2%, and Carbon ranked fourth on WebDev Arena. On the AA-Omniscience test, Carbon's accuracy rose to 58% from 48%, but its hallucination rate increased to 11% from 4%. Carbon is not yet broadly available; Google will first give access to select cybersecurity defenders and partners, with no general-release date announced. Pricing runs $2 per million input tokens and $10 per million output tokens under introductory rates.