Moonshot's New Model Closes AI Gap
The Chinese firm's latest release is narrowing the performance difference between open and proprietary AI systems.
By Adewale Fletcher ·
A new AI model from China is narrowing the gap between open and proprietary systems.
Moonshot AI has launched Kimi K3.1 Pro, a model that scored 55 points on the Artificial Analysis Intelligence Index. The release places the model at number four on the list, moving it up 11 points from its predecessor. According to the source, only Claude Opus 5.5 (58), Claude Sonnet 5.5 (56), and Muse Wonder Full (56) are currently ranked higher.
The company plans to release the model weights by November 4, 2026. If this schedule holds, K3.1 Pro would become the highest-scoring open-weights model available. Under those circumstances, the performance gap between open and proprietary models would drop from 12 points to just 3, assuming other models do not change their scores. Moonshot is also offering a variant named K3.1 Pro Highspeed, which the company states runs two to three times faster. Both versions are available now through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
The model shows increased capability in agentic tasks, coding, and 3D modeling. On the GDPval-AA benchmark, K3.1 Pro recorded an Elo of 1,846, which is 312 points higher than the previous K3 model. This score places it just behind Claude Opus 5.5 (1,866) and Muse Wonder Full (1,857). On Terminal-Bench 4.0, the model achieved 65.7 percent, surpassing GPT-6 Astra (58.2 percent) and nearly matching Claude Opus 5.5 (66.4 percent). By comparison, K3 scored only 13 percent on that test.
Hallucination rates also declined significantly. According to Moonshot's internal Kimi Hallu Bench Hard, claim-level hallucinations decreased from 23.4 percent with K3 to 8.3 percent with the new model. The company states this figure represents a stress test rather than a measure of general error rates. For financial tasks, K3.1 Pro gained 7.2 points on Financial Modeling and 11.6 points on Financial Research.
Regarding creative and coding capabilities, K3.1 Pro can work directly with game engines such as Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, the model scored 92.7 percent with tools, a result close to Claude Opus 5.5 (96.2 percent). Without tools, it reached 72.8 percent, nearly matching Claude Opus 5.5 (73.0 percent). Moonshot states the model is strongest at parametric modeling of mechanical parts, architectural models, and configurable products, and can write and run scripts for Blender and CAD software without MCP connectors.
Despite these capabilities, the model's architecture remains unchanged from its predecessor. K3.1 Pro retains 2.8 trillion total parameters and 104 billion active ones, with a context window spanning one million tokens. Artificial Analysis places the model at the top of the Pareto frontier, noting that no model with a comparable parameter count scores higher on the Intelligence Index. Token prices dropped by 33 percent. Input costs are now $2 per million tokens, down from $3, and output costs are $10, down from $15.
Moonshot has also adjusted the model's cybersecurity capabilities for the open-weights release. The company removed cyber-related data from both training phases to limit offensive capabilities. On its internal xdev-bench-hard, which reproduces recently disclosed vulnerabilities in software like the Linux kernel, Node.js, and JavaScriptCore, the uncleaned version solves 90.5 percent of tasks. After data cleaning, K3.1 Pro solves 81.0 percent.
