Artificial Analysis has released v5 of its Intelligence Index, a composite measure of AI capabilities that ranks Claude Opus 5.5 as the leading model with a score of 58. The updated index narrowly places GPT-6 Astra in second with 57 and GPT-6.1 Sol in third with 55, with Anthropic and OpenAI holding the top five spots. For the first time, 60% of the evaluations in the index are based on private test sets or held-out answers, removing saturated and public dataset questions to better mirror real-world usage.
The benchmark results show that the performance gap between proprietary and open weights models widened from 3 to 12 points, with Kimi K3.1 Pro leading the open category at 46. The company also noted that models are performing more work per task, with median output tokens rising from 35k to 59k. This increased demand is driven by agentic evaluations, which account for a median of 88% of output tokens. Additionally, Artificial Analysis launched the Progress Index v1 to track the capability of language models released since November 2022 on a continuous scale.
