Kimi K3.1 Pro ranks fourth on Artificial Analysis Intelligence Index

Artificial Analysis ranked Kimi K3.1 Pro fourth on its Intelligence Index, with a score of 55.[1] It said only Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full scored higher.[1] Kimi K3.1 scored 44, making the new model’s result an 11-point increase, according to Artificial Analysis.[1] The ranking offers professionals a comparison point, but the reported result is specific to this index; it does not by itself describe performance on every task or deployment.

Moonshot introduced Kimi K3.1 Pro as the first model in its K3.1 series and described it as its most capable 3T-class open-weight model to date.[2] The company said it is a post-trained version of Kimi K3, using substantially more post-training compute.[2] Moonshot also reported a one-million-token context window.[2] The model card lists 2.8 trillion total parameters and 104 billion activated parameters, and describes the model as natively multimodal.[3]

At launch, Moonshot said Kimi K3.1 Pro was available on Kimi.ai, Kimi Work, Kimi Code and the Kimi API.[2] It also released Kimi K3.1 Pro Highspeed, which the company said runs two to three times faster than K3.1 Pro.[2] Moonshot said it would publish the full model weights by November 4, 2026.[2] Artificial Analysis said the model would become the highest-scoring open-weights model once those weights were published.[1] It estimated that publication would narrow the Intelligence Index gap between leading open-weight and proprietary models from 12 points to three.[1]

Artificial Analysis ranked Kimi K3.1 Pro third on GDPval-AA, with 1,846 Elo—312 Elo above Kimi K3—and reported a 73% score on AutomationBench-AA, up from 58% for Kimi K3.[1] Moonshot reported a 65.7% score on another evaluation, compared with 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5.[2] It also reported a 61.2% score against Kimi K3’s 55.95%, and a 64.4 score on OfficeQA Pro v2, which it described as state of the art.[2] Moonshot said it removed cyber-related data from both training stages while retaining general-purpose coding tasks.[2]

Artificial Analysis reported Kimi K3.1 Pro API prices of $2 per million input tokens, $10 per million output tokens and $0.20 per million cache-hit tokens.[1] It said each rate was 33% below Kimi K3’s corresponding rate.[1] At the same time, its estimated cost per task was $1.98, roughly equal to Kimi K3.[1] Teams comparing options should keep token rates and estimated task cost distinct, and confirm which rates apply to the model or version they plan to use.