Chinese AI Model Threatens to Shrink Open vs. Closed Capability Gap

A new Chinese AI model is narrowing the performance gap with the industry's top proprietary systems.

Moonshot AI has rolled out Kimi K3.1 Pro, a post-trained iteration of its predecessor Kimi K3. The model achieved a score of 55 on the Artificial Analysis Intelligence Index, an improvement of 11 points over the previous version. This result places it at number four on the index, just three points behind the highest-ranking proprietary models.

Currently, the top three positions are held by Claude Opus 5.5 with a score of 58, followed by Claude Sonnet 5.5 and Muse Wonder Full, both at 56. According to Artificial Analysis, only those models outperform K3.1 Pro.

The model shows marked improvements in agentic tasks and coding. On the GDPval-AA benchmark, K3.1 Pro reached an Elo score of 1,846, an increase of 312 points over K3, placing it just behind Claude Opus 5.5 and Muse Wonder Full. On the Terminal-Bench 4.0 test, the model recorded a 65.7 percent success rate, narrowly trailing Claude Opus 5.5 at 66.4 percent but surpassing GPT-6 Astra at 58.2 percent. By comparison, the prior K3 model scored only 13 percent on that test. Hallucination rates also declined, dropping from 23.4 percent to 8.3 percent on Moonshot's internal Kimi Hallu Bench Hard. Moonshot describes this as a stress test rather than a general measure of error rates; GPT-6 Astra records 5.5 percent on the same benchmark.

The model also demonstrates capabilities in 3D modeling and financial analysis. On the BenchCAD benchmark, it scored 92.7 percent with tools enabled, close to the 96.2 percent of Claude Opus 5.5. Moonshot states the model is capable of writing and running scripts for Blender and CAD software without the need for MCP connectors. On internal finance benchmarks, K3.1 Pro gained 7.2 points on Financial Modeling and 11.6 points on Financial Research.

Despite these performance gains, the model's architecture remains unchanged from K3, retaining 2.8 trillion total parameters and 104 billion active parameters. It features a context window of one million tokens and supports both text and image inputs. Artificial Analysis reports that the model sits at the top of the Pareto frontier, noting that no model with a comparable parameter count scores higher on the Intelligence Index.

Pricing for the model has been adjusted downward. Token costs were reduced by 33 percent, with input priced at $2 per million tokens, down from $3, and output at $10, down from $15. The cost to run an Intelligence Index task is $1.98, nearly identical to the $2.00 cost for K3, as the model generates 14 percent more output tokens.

Moonshot has also limited the model's offensive capabilities. The company stated it removed cyber-related data from both training phases for the open-weights release. On the internal xdev-bench-hard test, the uncleaned version of the model solved 90.5 percent of tasks, while the cleaned version solves 81.0 percent. DeepSeek-V4.1-Flash reaches 85.7 percent on the same test.

Moonshot plans to release the model weights by November 4, 2026. If this schedule holds, K3.1 Pro will become the highest-scoring open-weights model available. This would reduce the gap between open and proprietary models to three points, assuming no other models improve their scores in the interim. A variant called K3.1 Pro Highspeed is also available, which Moonshot says runs two to three times faster. Both models can be accessed through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.