Gap Narrows to 3 Points as Moonshot AI Unveils K3.1 Pro

Chinese developer Moonshot AI has released Kimi K3.1 Pro, a reasoning model that immediately jumped to fourth place on the Artificial Analysis Intelligence Index. The model earned a score of 55, a jump of 11 points over its predecessor. It now trails only Claude Opus 5.5, which holds 58 points, along with Claude Sonnet 5.5 and Muse Wonder Full, both at 56. The release marks a significant step for open-weights AI, bringing the gap between open and proprietary models down to 3 points if the model weights are released as planned by November 4, 2026.

The model's architecture remains unchanged from K3, maintaining 2.8 trillion total parameters and 104 billion active parameters. Moonshot AI reports that the model is now available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. Alongside the main model, the company is also shipping a variant called K3.1 Pro Highspeed, which is designed to run two to three times faster than standard models.

The biggest gains for K3.1 Pro appear in agentic tasks, where the model learns to operate autonomously. On the GDPval-AA benchmark, the model reached 1,846 Elo, an increase of 312 points over the previous version. This places it just behind Claude Opus 5.5, which sits at 1,866. Performance on Terminal-Bench 4.0 also improved significantly, reaching 65.7 percent, compared to the 13 percent score of K3. On internal finance benchmarks, the model gained 7.2 points on Financial Modeling and 11.6 points on Financial Research.

Hallucination rates dropped sharply in testing. On the internal Kimi Hallu Bench Hard benchmark, claim-level hallucinations fell from 23.4 percent to 8.3 percent. Moonshot notes that this is a stress test rather than a measure of general error rates, and that GPT-6 Astra scores 5.5 percent on the same test. In other areas, the model tops all tested models on OfficeQA Pro v2, according to Moonshot.

The model shows specific capabilities in 3D modeling and coding, working directly with engines like Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, it scores 92.7 percent with tools, close to Claude Opus 5.5 at 96.2 percent and GPT-6 Astra at 95.9 percent. Moonshot states the model is strongest at parametric modeling of mechanical parts and architectural models. It can write and run scripts for Blender and CAD software without connectors, and the company has launched a feature called 3D Make on Kimi.com.

Pricing for the model has been reduced, with token costs dropping by 33 percent. Input costs $2 per million tokens, down from $3, while output costs $10, down from $15. However, each Intelligence Index task still costs $1.98, nearly identical to K3 at $2.00, because the model generates 14 percent more output tokens. Moonshot also stripped cyber-related data from the training phases for the open-weights release, which lowered its score on xdev-bench-hard from 90.5 percent to 81.0 percent.