Kimi K3.1 Pro cuts API costs by a third while scoring 55 points

The cost of running Artificial Analysis Intelligence Index tasks for Moonshot AI's new model is $1.98. This figure is nearly identical to the $2.00 cost for its predecessor, Kimi K3. The price stability comes despite a 33 percent reduction in raw API rates. Input tokens now cost $2 per million, output tokens $10, and cache hits $0.20.

K3.1 Pro is a post-trained version of K3 that received substantially more compute during the post-training phase. The model scored 55 points on the Artificial Analysis Intelligence Index. That score represents an 11-point improvement over K3. It places the model fourth on the index, behind Anthropic's Claude Opus 5.5 (58 points), Claude Sonnet 5.5 (56 points), and Muse Wonder Full (56 points).

The gap between the top proprietary models and the highest-scoring open-weights model will shrink from 12 points to 3 points. Moonshot plans to publish the model weights by November 4, 2026. If the schedule holds, K3.1 Pro will become the highest-scoring open-weights model available. Artificial Analysis places the model on the top of the Pareto frontier, meaning no model of comparable size scores higher on its index.

Performance gains are most visible in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro reached 1,846 Elo. That is 312 points above K3 and trails only Claude Opus 5.5 (1,866) and Muse Wonder Full (1,857). On Terminal-Bench 4.0, the new model scored 65.7 percent. K3 managed only 13 percent on that test. GPT-6 Astra scored 58.2 percent, while Claude Opus 5.5 led with 66.4 percent.

Moonshot reports a sharp drop in hallucinations. On its internal Kimi Hallu Bench Hard, claim-level hallucinations fell from 23.4 percent with K3 to 8.3 percent. The company describes this benchmark as a stress test rather than a measure of general error rates. GPT-6 Astra still leads this metric at 5.5 percent. On internal finance benchmarks, K3.1 Pro gained 7.2 points on Financial Modeling and 11.6 points on Financial Research. Moonshot states that the model tops all tested systems on OfficeQA Pro v2.

The model supports 3D modeling and coding for game engines including Unity, Unreal Engine, and Godot. It can write and run scripts for Blender and CAD software without MCP connectors. On the BenchCAD benchmark, K3.1 Pro scores 92.7 percent with tools. That compares to 96.2 percent for Claude Opus 5.5 and 95.9 percent for GPT-6 Astra. Without tools, the score is 72.8 percent, slightly below Claude Opus 5.5 at 73.0 percent. A new 3D Make feature on Kimi.com allows users to preview, edit, and export models.

The architecture remains unchanged from K3. The model has 2.8 trillion total parameters, with 104 billion active. It supports a one-million-token context window and handles both text and images. Alongside the main model, Moonshot offers K3.1 Pro Highspeed. The company says this variant runs two to three times faster. Both versions are available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.

Moonshot devotes a section of its technical blog to cybersecurity. Vulnerability research capabilities grew as post-training scaled up because cyber-related content was part of the general mid-training corpus. For the open-weights release, the company removed cyber-related data from both training phases. On its internal xdev-bench-hard, which reproduces recently disclosed vulnerabilities in software like the Linux kernel and Node.js, the uncleaned model solved 90.5 percent of tasks. After cleaning, that figure dropped to 81.0 percent. DeepSeek-V4.1-Flash scored 85.7 percent, while Kimi K3 scored 33.3 percent.

Moonshot states that no open-weights model should push the absolute frontier of offensive cyber capabilities. The company aims to raise what open models can do gradually to support defensive work such as auditing legacy systems. Anthropic previously showed that GLM-5.3, which scores 45 on the Artificial Analysis index, can write exploits at levels close to Claude Mythos.