Chinese AI model surges to fourth place

The new model from Moonshot AI scores 55 points on the Artificial Analysis Intelligence Index. This result places Kimi K3.1 Pro in fourth position globally. It sits just three points behind the current leaders. The closest rivals are Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full. This marks a significant shift in the competitive landscape. The previous version, Kimi K3, scored 44 points. The jump to 55 represents an 11-point increase. The company used significantly more post-training compute to achieve this. If weights are released by November 4, 2026, it becomes the top open-weights model. This would shrink the performance gap to just three points.

The improvements are most visible in agentic tasks. On the GDPval-AA benchmark, the model reaches 1,846 Elo. This is 312 points higher than its predecessor. It trails only Claude Opus 5.5 and Muse Wonder Full in this category. Coding performance also saw a major leap. Moonshot reports a 65.7 percent score on Terminal-Bench 4.0. The old K3 model managed only 13 percent on that same test. GPT-6 Astra scored 58.2 percent, placing it behind the new release. Claude Opus 5.5 leads with 66.4 percent. Hallucination rates dropped sharply as well. Internal tests show claim-level errors falling from 23.4 percent to 8.3 percent. Moonshot notes this is a stress test, not a general measure. GPT-6 Astra still holds an edge at 5.5 percent.

The model now works directly with game engines like Unity and Unreal. It handles quick fixes and full project builds. On BenchCAD, it scores 92.7 percent with tools. This is close to Claude Opus 5.5 at 96.2 percent. Without tools, it hits 72.8 percent, nearly matching the leader. A new feature called 3D Make allows users to preview and export models. The company says it excels at parametric modeling for mechanical parts. It can write scripts for Blender and CAD software without extra connectors. Finance benchmarks also improved, with gains in modeling and research tasks.

The architecture remains unchanged from the previous version. It maintains 2.8 trillion total parameters and 104 billion active ones. The context window supports one million tokens. It processes both text and images. Prices have dropped by 33 percent to make it more accessible. Input costs are now $2 per million tokens. Output costs are $10, and cache hits cost $0.20. The cost per task on the Intelligence Index is $1.98. This is almost identical to the previous model. The price cut reflects a 14 percent increase in output tokens.

Moonshot addressed cybersecurity concerns in its technical blog. Vulnerability research capabilities grew as post-training scaled up. Cyber-related content was embedded in the training data. For the open-weights release, the company stripped this data. On an internal test, the uncleaned version solved 90.5 percent of tasks. After cleaning, K3.1 Pro solved 81.0 percent. DeepSeek-V4.1-Flash reached 85.7 percent on the same test. The old Kimi K3 scored 33.3 percent. Moonshot aims to balance offensive and defensive capabilities. They want to gradually raise the standard for open models. A high-scoring model can still write effective exploits.

Both K3.1 Pro and the faster Highspeed variant are available now. They run through Kimi.ai, Kimi Work, and Kimi Code. The Highspeed version runs two to three times faster. Users can access the models via the API today. The weights are scheduled for release next month. This timing could reshape the open-source AI market. The gap between proprietary and open models will narrow. Developers in the Pacific Northwest and beyond can test the tools immediately. The lower price point makes it attractive for enterprise use. The next few months will determine its long-term impact. The race to the top continues with this new entry.