The competitive landscape of large models has been reshuffled again. According to the latest released Artificial Analysis Intelligence Index, Anthropic's flagship model Claude Opus4.6 has officially topped the authoritative ranking with outstanding performance. This index comprehensively evaluates ten deep tests including programming, agent tasks, and scientific reasoning. Opus4.6 secured first place in agent task work, terminal programming, and physics research topics.

image.png

Notably, despite the operational cost of Opus4.6 reaching $2,486, slightly higher than OpenAI's GPT-5.2 at $2,304, there is a significant difference in efficiency performance. Data shows that Opus4.6 consumed approximately 58 million output tokens during testing. Although this number is twice that of its predecessor, version 4.5, it is significantly more efficient compared to GPT-5.2's staggering 130 million tokens. Currently, the model is fully available on the Claude.ai platform and supports invocation through mainstream cloud services such as Google Vertex and AWS Bedrock.

However, Anthropic's leading position is facing serious challenges. Industry giant OpenAI's new programming tool Codex5.3 is already on the test list. Analysts point out that once Codex5.3 completes all benchmark tests, its advantages in code writing and related logic fields are likely to help it reclaim the top spot. The battle for the "smartest model in the world" is far from over.