Moonshot AI Narrows Gap With Leading US Models

The competition for artificial intelligence supremacy in North America has intensified as Chinese developers close the distance to American leaders. Moonshot AI released Kimi K3.1 Pro on Thursday, a new reasoning model that immediately claimed the fourth position on the Artificial Analysis Intelligence Index. This ranking places it within striking distance of the top proprietary systems currently dominating the market.

The new model scored 55 points on the index, marking an 11 point jump over its predecessor, Kimi K3. It now sits just three points behind Claude Opus 5.5, which holds the lead with 58 points. Claude Sonnet 5.5 and Muse Wonder Full also rank higher with 56 points each. If Moonshot releases the model weights by November 4, 2026, as planned, K3.1 Pro would become the highest scoring open weights model available. This move would shrink the gap between open and proprietary models from 12 to just 3 points, assuming no other changes occur by then.

Significant improvements are evident in agentic tasks, where the model shows its strongest gains. On the GDPval-AA benchmark, K3.1 Pro reached 1,846 Elo, a rise of 312 points over K3. This places it narrowly behind Claude Opus 5.5 and Muse Wonder Full. In terminal operations, Moonshot reports a score of 65.7 percent on Terminal-Bench 4.0. This result surpasses GPT-6 Astra and trails Claude Opus 5.5 by less than a single point. The previous version, K3, had managed only 13 percent on the same test.

Reliability has also improved, with hallucination rates dropping sharply. On Moonshot's internal Kimi Hallu Bench Hard, claim level hallucinations fell from 23.4 percent with K3 to 8.3 percent. The company notes this is a stress test rather than a measure of general error rates. GPT-6 Astra still holds an edge on this benchmark with a 5.5 percent rate. In finance, the new model gained 7.2 points on Financial Modeling and 11.6 points on Financial Research. Moonshot also claims it tops all tested models on OfficeQA Pro v2 for general office tasks.

The model demonstrates strong capabilities in 3D modeling and coding, working directly with game engines such as Unity, Unreal Engine, and Godot. It handles both quick fixes and full project builds. On the BenchCAD benchmark, it scores 92.7 percent with tools, close to Claude Opus 5.5 and GPT-6 Astra. Without tools, it hits 72.8 percent, nearly matching the leading US model. Moonshot states the system is strongest at parametric modeling of mechanical parts and architectural designs. It can write and run scripts for Blender and CAD software without MCP connectors.

Despite the performance boost, the physical size of the model remains unchanged. K3.1 Pro keeps the same dimensions as K3, with 2.8 trillion total parameters and 104 billion active ones. The context window spans one million tokens, and the model handles both text and images. According to Artificial Analysis, it sits at the top of the Pareto frontier, meaning no model with a comparable parameter count scores higher on the Intelligence Index.

API prices have dropped by 33 percent alongside the performance increase. Input costs now stand at $2 per million tokens, down from $3. Output costs are $10, reduced from $15, and cache hits cost $0.20. Each Intelligence Index task costs $1.98, which is almost identical to the previous cost since the model generates 14 percent more output tokens. Moonshot is also shipping a variant called K3.1 Pro Highspeed that runs two to three times faster.

The company devoted a section of its technical blog to cybersecurity, noting that vulnerability research capabilities grew as post-training scaled up. For the open weights release, Moonshot stripped cyber-related data from both training phases. On its internal xdev-bench-hard, the uncleaned version solves 90.5 percent of tasks, while the cleaned K3.1 Pro solves 81.0 percent. DeepSeek-V4.1-Flash reaches 85.7 percent on the same test. Moonshot argues that no open weights model should push the absolute frontier of offensive cyber capabilities, but it wants to gradually raise what open models can do for defensive purposes.