55 Points: Chinese AI Model Kimi K3.1 Pro Narrows Gap With Claude Opus

Moonshot AI has released Kimi K3.1 Pro, a new reasoning model that has surged to fourth place on the Artificial Analysis Intelligence Index with a score of 55. This marks an 11-point improvement over its predecessor, Kimi K3, and places the model within three points of the top proprietary contenders. The release signals a continued narrowing of the performance gap between Chinese-developed open-weights models and leading American proprietary systems.

The model sits just behind Claude Opus 5.5, which holds the top spot with 58 points. It also trails Claude Sonnet 5.5 and Muse Wonder Full, both of which score 56 points. If Moonshot adheres to its plan to release the model weights by November 4, 2026, Kimi K3.1 Pro would become the highest-scoring open-weights model available. This milestone would reduce the performance disparity between the best open and proprietary models from 12 points to just three, assuming no other major releases occur before then.

Significant performance gains are most evident in agentic tasks. On the GDPval-AA benchmark, Kimi K3.1 Pro achieved an Elo rating of 1,846, a 312-point increase over Kimi K3. This result places it closely behind Claude Opus 5.5 (1,866) and Muse Wonder Full (1,857). In coding assessments, the model reported a 65.7 percent score on Terminal-Bench 4.0, surpassing GPT-6 Astra’s 58.2 percent and coming within 0.7 percent of Claude Opus 5.5’s 66.4 percent. By comparison, the previous Kimi K3 model scored only 13 percent on the same test.

Hallucination rates have also improved substantially. According to Moonshot’s internal Kimi Hallu Bench Hard, claim-level hallucinations dropped from 23.4 percent in the previous version to 8.3 percent in K3.1 Pro. The company notes this is a stress test rather than a measure of general error rates. GPT-6 Astra maintains a lead with a 5.5 percent hallucination rate on this specific benchmark. In financial tasks, K3.1 Pro saw gains of 7.2 points in Financial Modeling and 11.6 points in Financial Research, while topping all tested models on OfficeQA Pro v2 for general office tasks.

The model demonstrates strong capabilities in 3D modeling and game development, working directly with engines such as Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, it scored 92.7 percent when using tools, slightly behind Claude Opus 5.5 (96.2 percent) and GPT-6 Astra (95.9 percent). Without external tools, it achieved 72.8 percent, nearly matching Claude Opus 5.5’s 73.0 percent. Moonshot states the model is particularly effective at parametric modeling of mechanical parts and architectural structures, capable of writing and running scripts for Blender and CAD software without MCP connectors.

Kimi K3.1 Pro retains the same architectural dimensions as Kimi K3, featuring 2.8 trillion total parameters with 104 billion active parameters. It supports a context window of one million tokens and processes both text and images. Artificial Analysis indicates the model sits at the top of the Pareto frontier, meaning no model with a comparable parameter count scores higher on the Intelligence Index. A high-speed variant, K3.1 Pro Highspeed, is also available, running two to three times faster according to the company. Both versions are accessible via Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.

Despite the larger scale of post-training, the API prices have decreased by 33 percent. Input costs are now $2 per million tokens, down from $3, while output costs have fallen to $10 from $15. Cache hits are priced at $0.20, a reduction from $0.30. The total cost per Intelligence Index task is $1.98, nearly identical to the previous model’s $2.00, primarily because the new model generates 14 percent more output tokens.

Moonshot has also outlined specific measures regarding cybersecurity capabilities. The company observed that vulnerability research skills improved as post-training compute increased. To prepare for the open-weights release, cyber-related data was stripped from both training phases. On the 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 data cleaning, K3.1 Pro solved 81.0 percent, compared to 85.7 percent for DeepSeek-V4.1-Flash and 33.3 percent for Kimi K3. Moonshot aims to balance offensive capabilities with defensive utility for legacy system auditing.