Moonshot AI’s New Model Narrows the Gap With Top Rivals

A major Chinese artificial intelligence firm has released a new reasoning model that significantly closes the performance gap with leading global systems. Moonshot AI launched Kimi K3.1 Pro, which now holds the fourth position on the Artificial Analysis Intelligence Index.

The new model achieved a score of 55 points, marking an 11 point jump over its predecessor, Kimi K3. This result places it only three points behind the highest scoring proprietary systems. Currently, only Anthropic’s Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full rank higher on the index. Moonshot intends to release the model weights by November 4, 2026. If that release occurs as planned, K3.1 Pro would become the top performing open weights model available, shrinking the difference between open and proprietary options from 12 points to just three.

In addition to the standard release, the company introduced a variant called K3.1 Pro Highspeed. This version is designed to run two to three times faster than the main model. Both options are accessible through Kimi.ai, Kimi Work, Kimi Code, and the company’s API platform.

Performance gains were most evident in agentic tasks. On the GDPval AA benchmark, the model reached 1,846 Elo, surpassing K3 by 312 points. It trails only Claude Opus 5.5 and Muse Wonder Full in this specific metric. In coding evaluations, K3.1 Pro scored 65.7 percent on Terminal Bench 4.0. This result exceeds GPT 6 Astra at 58.2 percent and sits slightly below Claude Opus 5.5 at 66.4 percent. By comparison, the older K3 model managed only 13 percent on that same test.

The update also addresses a common issue with large language models: hallucinations. Internal testing on the Kimi Hallu Bench Hard shows claim level hallucinations dropped from 23.4 percent in K3 to 8.3 percent in the new version. While GPT 6 Astra still leads with a 5.5 percent error rate, Moonshot notes the benchmark serves as a stress test rather than a measure of general error. The model also improved in financial analysis, gaining 7.2 points in modeling and 11.6 points in research. It reportedly tops all tested models on OfficeQA Pro v2.

Capabilities in 3D modeling and coding expanded as well. K3.1 Pro integrates with game engines like Unity, Unreal Engine, and Godot. It can also write and execute scripts for Blender and CAD software without requiring MCP connectors. On the BenchCAD benchmark, the model scored 92.7 percent with tools enabled. This compares to 96.2 percent for Claude Opus 5.5 and 95.9 percent for GPT 6 Astra. Without tools, it reached 72.8 percent, nearly matching Claude Opus 5.5 at 73.0 percent. A new 3D Make feature on Kimi.com allows users to preview, edit, and export 3D models.

The underlying architecture remains consistent with K3, featuring 2.8 trillion total parameters with 104 billion active. It supports a one million token context window for text and images. Artificial Analysis places the model on the Pareto frontier, indicating no model of similar size scores higher. API prices decreased by 33 percent, with input costs at $2 per million tokens and output at $10. Cache hits cost $0.20. Despite generating 14 percent more output tokens, the cost per Intelligence Index task is $1.98, almost identical to the previous $2.00.

Moonshot provided details on cybersecurity capabilities in its technical blog. Vulnerability research improved as post training scaled up, aided by cyber content in the general mid training corpus. For the open weights release, the company removed such data from both training phases. On its internal xdev bench hard test, which reproduces recent vulnerabilities in software like the Linux kernel, the uncleaned model solved 90.5 percent of tasks. After cleaning, this figure dropped to 81.0 percent. DeepSeek V4.1 Flash scored 85.7 percent, while Kimi K3 managed 33.3 percent. The company aims to gradually raise open model capabilities for defensive work, noting that even lower scoring models can write exploits comparable to top proprietary systems.