Artificial analysis launches intelligence index v5, DeepSeek V4.1 flash 1017 leads open-source benchmark

Artificial Analysis released the Intelligence Index v5 on October 28, 2026, marking the most significant benchmark overhaul since its v4 launch in January 2026. The update shifts 60% of the evaluation weight to private test sets and hidden answers, reducing the influence of public datasets and minimizing score inflation. The index now includes nine categories, with FrontierCode 2.0, a new coding benchmark, contributing 15% of the coding weight. DeepSeek V4.1 Flash 1017 achieved the highest score of 28% on FrontierCode 2.0, surpassing other models like Kimi K3.1 Pro and Gemini 4 Argon. The v5 index highlights a growing gap between open-source and closed-source models, with the lead now at 12 points compared to 3 points in v4.3. While Kimi K3.1 Pro leads in open-source models, its weights are not yet released. The index also emphasizes the cost differences between models, with some showing performance gaps of over 10 times depending on effort levels. Progress Index v1 was introduced to compare models across different benchmark versions, mapping capabilities from November 2022 onwards.