Google’s Gemini 4 Carbon Tops Artificial Analysis Intelligence Index, Becomes Top AI Model In The World

Google's Gemini 4 Carbon took the #1 spot on the Artificial Analysis Intelligence Index v5.0 with a score of 59, edging Claude Opus 5.5 at 58 and GPT-6 Astra at 57, according to the benchmark release. The model also posted 70.4% on the Vals Index v2.0, 65.2% on Terminal-Bench 4.0, 77.4% on Agents' Last Exam v1.1, 81.5% on the OSWorld offline subset, 80.0% on RiemannBench and 88.2% on SEC-bench Pro, with long-context reasoning scaling from 256k to 1M tokens. The ranking restores Google to the top of the frontier-model table as Artificial Analysis v5.0 adds private test sets and agentic environments to its evaluation standards.

After nearly a year in the wilderness, Google is back to the top of the AI pile. Google has officially seized the crown in the global artificial intelligence race. With the rollout of its next-generation frontier model, Gemini 4 Carbon , Google has surged to the 1 spot on the newly updated Artificial Analysis Intelligence Index v5.0 , scoring an impressive 59 points . The latest benchmark release highlights a major shift in the AI landscape, with Gemini 4 Carbon leapfrogging competitors like Claude Opus 5.5 scoring 58 and GPT-6 Astra scoring 57 to claim the title of the world’s most intelligent AI model. A Comprehensive Sweep Across Benchmarks Google’s latest model didn’t just win by a narrow margin; it demonstrated a dominant, cross-domain performance advantage. According to benchmark https://officechai.com/ai/gemini-4-carbon-benchmarks/ metrics accompanying the v5.0 index release, Gemini 4 Carbon excels across multiple technical and enterprise vectors: - General Capabilities: Gemini 4 Carbon leads the pack with a score of 59 on the AA Index v5.0 and a stellar 70.4% on the Vals Index v2.0. - Agentic Coding & Terminal Tasks: Proving its value to software engineering teams, Carbon scored 65.2% on Terminal-Bench 4.0 and strong figures in agentic coding frameworks. - Computer Use & Knowledge Work: In practical enterprise scenarios, the model achieved 77.4% on Agents’ Last Exam v1.1 and 81.5% on the OSWorld offline subset, alongside handling complex workflows like Vending-Bench 2 $16,484 final bank balance . - Science, Math, and Cybersecurity: Gemini 4 Carbon scored 80.0% on RiemannBench and delivered groundbreaking capabilities in cybersecurity defense, hitting an unprecedented 88.2% on SEC-bench Pro to autonomously uncover complex flaws in real-world software. What This Means for Enterprise and AI As AI shifts rapidly from simple text-generation chat interfaces to autonomous agents capable of executing multi-step computer tasks, coding, and cyber-defense, raw intelligence metrics matter more than ever. Google’s design of Gemini 4 Carbon as a natively multimodal, reasoning powerhouse appears to have paid off, giving enterprises a tool capable of deep visual understanding, long-context reasoning scaling smoothly from 256k to 1M tokens , and complex multi-turn investigations. With Artificial Analysis tightening its evaluation standards—incorporating rigorous private test sets and demanding agentic environments in Index v5.0—Gemini 4 Carbon’s top-tier ranking solidifies Google’s leadership position in the current frontier of AI development.