Moonshot AI releases Kimi K3.1 Pro with lower costs and top agentic scores

Moonshot AI has launched Kimi K3.1 Pro, a reasoning model that currently ranks fourth on the Artificial Analysis Intelligence Index. The system scored 55 points, marking an 11 point increase over the previous version, Kimi K3. It trails only Anthropic's Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full.

The company reduced API prices by 33 percent with the release. Input tokens cost $2 per million, output tokens cost $10 per million, and cache hits cost $0.20 per million. Because the model generates 14 percent more output tokens, the total cost to run each Intelligence Index task is $1.98. This figure is nearly identical to the $2.00 cost associated with the previous model.

Moonshot plans to publish the model weights by November 4, 2026. If released on schedule, K3.1 Pro would become the highest scoring open weights model available. That move would shrink the gap between open and proprietary systems from 12 points to three points on the Artificial Analysis index.

Performance gains were most evident in agentic tasks. On the GDPval-AA benchmark, the model achieved 1,846 Elo points. That score is 312 points higher than K3 and sits just behind Claude Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. On Terminal-Bench 4.0, Moonshot reported a score of 65.7 percent. This result exceeds GPT-6 Astra's 58.2 percent and falls slightly short of Claude Opus 5.5's 66.4 percent. The previous Kimi K3 model managed only 13 percent on that specific test.

The company also reported a decrease in hallucinations. On its internal Kimi Hallu Bench Hard, claim-level hallucinations dropped from 23.4 percent with K3 to 8.3 percent with the new release. Moonshot describes this benchmark as a stress test rather than a general error rate measure. It notes that GPT-6 Astra still leads this metric with a 5.5 percent rate.

Financial benchmarks showed specific improvements as well. The model gained 7.2 points on Financial Modeling and 11.6 points on Financial Research. Moonshot stated that K3.1 Pro tops all tested models on OfficeQA Pro v2. The company also highlighted capabilities in 3D modeling and coding. The system works with game engines including Unity, Unreal Engine, and Godot. It can also write and run scripts for Blender and CAD software without MCP connectors.

On the BenchCAD benchmark, the model scored 92.7 percent with tools enabled. Claude Opus 5.5 recorded 96.2 percent and GPT-6 Astra recorded 95.9 percent on the same test. Without tools, K3.1 Pro reached 72.8 percent, compared to 73.0 percent for Claude Opus 5.5. A new 3D Make feature on Kimi.com allows users to preview, edit, and export 3D models.

The architecture remains unchanged from K3. The model features 2.8 trillion total parameters, with 104 billion active. It supports a one million token context window and accepts text and images. Artificial Analysis places the system at the top of the Pareto frontier, indicating no model of comparable size scores higher on its index.

Moonshot addressed cybersecurity capabilities in its technical blog. The company said vulnerability research improved as post-training scaled. Cyber-related content was part of the general mid-training corpus and included in the reinforcement learning mix. For the open weights release, the team removed cyber-related data from both training phases. On its internal xdev-bench-hard test, 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 and Kimi K3 scored 33.3 percent.

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