Moonshot Cuts AI Costs by Third

Moonshot AI released Kimi K3.1 Pro on October 23, 2026, dropping its API input price to $2 per million tokens and output to $10. The 33 percent reduction in rates makes the model more competitive for developers. The new system also scores 55 points on the Artificial Analysis Intelligence Index. This places it fourth among global models.

The score represents an 11 point increase over the previous Kimi K3 version. Only three systems currently outperform it. Anthropic's Claude Opus 5.5 leads with 58 points. Claude Sonnet 5.5 and Muse Wonder Full both hold 56 points. Moonshot plans to publish the model weights by November 4, 2026. If released as scheduled, K3.1 Pro would become the highest scoring open weights model available.

Performance gains are most visible in agentic tasks. On the GDPval-AA benchmark, the model reached 1,846 Elo. That is 312 points higher than K3. It trails only Claude Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. On Terminal-Bench 4.0, Moonshot reports a 65.7 percent success rate. This beats GPT-6 Astra at 58.2 percent. K3 previously managed just 13 percent on that specific test.

The company also reports a sharp decline in hallucinations. Claim level errors on the internal Kimi Hallu Bench Hard dropped from 23.4 percent to 8.3 percent. GPT-6 Astra still leads this metric at 5.5 percent. Finance capabilities improved as well. The model gained 7.2 points on Financial Modeling and 11.6 on Financial Research. Moonshot states it now tops all tested models on OfficeQA Pro v2.

Coding and 3D modeling capabilities expanded alongside the core improvements. K3.1 Pro integrates with Unity, Unreal Engine, and Godot. It writes scripts for Blender and CAD software without needing MCP connectors. On BenchCAD with tools, the model scores 92.7 percent. Claude Opus 5.5 leads that category at 96.2 percent. A new 3D Make feature on Kimi.com allows users to preview and export models directly.

The architecture remains identical to K3, featuring 2.8 trillion total parameters and 104 billion active parameters. The context window supports one million tokens for text and images. Artificial Analysis places the model on the top of the Pareto frontier. No model of comparable size scores higher on the index. The cost to run each Intelligence Index task is $1.98, nearly matching K3's $2.00.

Moonshot addressed cybersecurity concerns in its technical documentation. The company removed cyber related data from the training set for the open weights release. The uncleaned model solved 90.5 percent of tasks on the xdev-bench-hard test. The cleaned version solved 81.0 percent. DeepSeek-V4.1-Flash scored 85.7 percent on the same benchmark. Moonshot says open models should raise the bar for offensive capabilities gradually while supporting defensive work.