Moonshot AI’s Kimi K3.1 Pro moves closer to the leading proprietary models on the Artificial Analysis Intelligence Index, while cutting API prices. The company describes it as a post-trained version of Kimi K3, with the largest reported gains in agentic tasks, coding, and 3D modeling.
A higher score, with a small gap to the leaders
Kimi K3.1 Pro scored 55 points on the Artificial Analysis Intelligence Index, an 11-point improvement over K3. It ranks fourth, behind Claude Opus 5.5 at 58, Claude Sonnet 5.5 at 56, and Muse Wonder Full at 56.
Moonshot says it plans to release the model weights by November 4, 2026. If that release happens as planned, and rankings stay otherwise unchanged, K3.1 Pro would become the highest-scoring open-weights model available. The source estimates that this would narrow the gap between open and proprietary models from 12 points to three.
The company is also offering a K3.1 Pro Highspeed variant, which it says runs two to three times faster. Both versions are available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Agentic work shows the biggest gains
On GDPval-AA, K3.1 Pro scores 1,846 Elo, which Artificial Analysis says is 312 points above K3. That places it just behind Claude Opus 5.5 at 1,866 and Muse Wonder Full at 1,857.
Moonshot reports a 65.7 percent result on Terminal-Bench 4.0. That is ahead of GPT-6 Astra at 58.2 percent and close to Claude Opus 5.5 at 66.4 percent. K3 scored 13 percent on the same test.
The company also reports a lower claim-level hallucination rate on its internal Kimi Hallu Bench Hard: 8.3 percent for K3.1 Pro, compared with 23.4 percent for K3. Moonshot cautions that this benchmark is a stress test, not a general measure of error rates. GPT-6 Astra scores 5.5 percent on it.
On internal finance benchmarks, K3.1 Pro gained 7.2 points in Financial Modeling and 11.6 points in Financial Research. Moonshot also says it tops all tested models on OfficeQA Pro v2, a benchmark for general office tasks.
Building and coding in 3D environments
Moonshot says K3.1 Pro can work directly with Unity, Unreal Engine, and Godot, from quick fixes to full project builds. On BenchCAD, it scores 92.7 percent with tools, compared with 96.2 percent for Claude Opus 5.5 and 95.9 percent for GPT-6 Astra. Without tools, its score is 72.8 percent, close to Claude Opus 5.5 at 73.0 percent.
The company says the model is particularly capable at parametric modeling for mechanical parts, architectural models, and configurable products. It can write and run scripts for Blender and CAD software without MCP connectors. Moonshot has also introduced “3D Make” on Kimi.com, where users can preview, edit, and export 3D models.
Same model size, lower API prices
K3.1 Pro retains K3’s dimensions: 2.8 trillion total parameters, with 104 billion active. It accepts text and images and has a one million token context window. Artificial Analysis places it at the top of the Pareto frontier, meaning no model with a comparable parameter count scores higher on the Intelligence Index, according to the source.
Moonshot has reduced token prices by 33 percent. Input costs $2 per million tokens, down from $3; output costs $10, down from $15; and cache hits cost $0.20, down from $0.30. Despite the lower rates, the reported cost per Intelligence Index task is $1.98, almost the same as K3’s $2.00, because K3.1 Pro generates 14 percent more output tokens.
Cybersecurity controls for open weights
Moonshot says it found that vulnerability research capabilities grew as post-training scaled up. The company linked that growth to cyber-related material in the general mid-training corpus and related tasks in its reinforcement learning environment mix.
For the planned open-weights release, Moonshot says it removed cyber-related data from both training phases. On the company’s xdev-bench-hard test, which reproduces recently disclosed vulnerabilities in the Linux kernel, Node.js, and JavaScriptCore, the uncleaned model solves 90.5 percent of tasks. K3.1 Pro solves 81.0 percent after data cleaning; DeepSeek-V4.1-Flash scores 85.7 percent, while Kimi K3 scored 33.3 percent.
Moonshot says open-weights models should not push the absolute frontier of offensive cyber capabilities. It also argues that raising their capabilities gradually can support defensive uses, such as auditing legacy systems. That balance is part of the model’s release story alongside its benchmark gains: the company is aiming for a stronger general-purpose system while setting limits on its offensive cybersecurity performance.