AI Benchmark Shifts: Moonshot's New Model Narrows Gap With US Rivals

If you rely on artificial intelligence tools for work or personal projects, the cost and performance landscape just got a little more competitive. A new model from a Chinese company has arrived that offers strong capabilities at a lower price point than some of its top rivals. This development matters because it could make advanced AI more accessible to developers and businesses that have been paying premium rates for the best available options.

Moonshot AI has launched Kimi K3.1 Pro, a significant upgrade to its previous version. The model immediately jumped to fourth place on the Artificial Analysis Intelligence Index, scoring 55 points. This places it just three points behind the leading proprietary models, which include Claude Opus 5.5 and Muse Wonder Full. The jump represents an 11 point increase over the older Kimi K3 version, showing substantial gains in reasoning and task completion.

The most notable improvements are seen in agentic tasks, which involve the AI acting independently to achieve goals. On the GDPval-AA benchmark, the new model reached an Elo score of 1,846, a huge leap of 312 points from its predecessor. It also performed strongly in coding tests, achieving a 65.7 percent score on Terminal-Bench 4.0. That result puts it ahead of GPT-6 Astra and very close to Claude Opus 5.5, which scored 66.4 percent. The previous version of the model only managed 13 percent on that same test.

Accuracy has also improved. Moonshot reported that hallucination rates dropped sharply on their internal stress test. Claim level errors fell from 23.4 percent in the old model to 8.3 percent in the new one. While this is a good sign, GPT-6 Astra still holds an edge with a 5.5 percent error rate on the same benchmark. The company notes that this specific test is designed to be difficult, so it may not reflect the model's general error rate in everyday use.

Beyond text and code, the model shows promise in 3D modeling. It can work directly with game engines like Unity and Unreal Engine, handling both small fixes and full project builds. On the BenchCAD benchmark, it scored 92.7 percent when using tools, which is close to the top performers. Moonshot says the model is particularly strong at parametric modeling for mechanical parts and architecture. It can even write and run scripts for software like Blender without needing special connectors.

Pricing is another key factor for users. The model keeps the same architecture as its predecessor, with 2.8 trillion total parameters and 104 billion active ones. However, API prices have dropped by 33 percent. Input tokens now cost $2 per million, down from $3, while output tokens are $10 per million, down from $15. Cache hits are also cheaper at $0.20 per million tokens. For most tasks, the cost remains nearly identical to the previous version because the new model generates slightly more output tokens.

Moonshot plans to release the model weights by November 4, 2026. If that happens, Kimi K3.1 Pro would become the highest scoring open weights model available. This would shrink the gap between open and proprietary models from 12 points to just 3 points. This move is significant for the industry because open weights models can be modified and run privately, which is a major draw for many developers and enterprises.

The company also took specific steps regarding cybersecurity capabilities. They observed that vulnerability research skills improved as training scaled up. For the open weights release, they removed cyber related data from the training process to prevent the model from being too powerful in offensive tasks. On an internal test involving known software vulnerabilities, the uncleaned version solved 90.5 percent of tasks. After cleaning the data, the score dropped to 81.0 percent. This is higher than some rivals but lower than the uncleaned version, striking a balance between utility and safety.

There are still open questions about how this model will perform in real world scenarios over time. While the benchmarks look impressive, independent verification and long term usage will determine its true value. The entry of a highly competitive Chinese model at a lower price point puts pressure on US based companies to maintain their lead. For now, users have a new option that offers strong performance at a more affordable cost.