Chinese AI Rival Cuts Price, Closes Gap With Top U.S. Models

The gap between China’s best AI models and the U.S. leaders is shrinking fast. Moonshot AI just released Kimi K3.1 Pro, a model that scores 55 points on the Artificial Analysis Intelligence Index. This puts it in fourth place, just three points behind the current champion, Claude Opus 5.5. For a region that has long trailed Silicon Valley, this is a significant leap forward. The model is now available through various platforms, including the Kimi API and Kimi Work.

This release does not happen in a vacuum. It follows a steady climb for Moonshot AI. The previous version, Kimi K3, scored 44 points. The jump to 55 points represents an 11-point gain. If Moonshot follows through on its plan to release the model weights by November 4, 2026, K3.1 Pro will become the highest-scoring open-weights model available. That would narrow the distance between open and proprietary models from 12 points to just three.

The new model shows its strongest gains in agentic tasks. On the GDPval-AA benchmark, it reaches 1,846 Elo, a massive 312-point increase over its predecessor. It sits just behind Claude Opus 5.5 and Muse Wonder Full. Coding performance has also improved dramatically. On Terminal-Bench 4.0, the model achieves a 65.7 percent success rate. This far surpasses GPT-6 Astra and nearly matches the top U.S. model. In previous tests, the older K3 version managed only 13 percent.

Reliability has improved as well. Hallucination rates dropped sharply. On Moonshot’s internal Kimi Hallu Bench Hard, claim-level hallucinations fell from 23.4 percent to 8.3 percent. The company notes this is a stress test, but the reduction is notable. In finance, the model gained points in both modeling and research tasks. For general office work, Moonshot claims the model tops all tested competitors on the OfficeQA Pro v2 benchmark.

The model also excels in 3D modeling and coding for game engines. It works directly with Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, it scores 92.7 percent when using tools. This is close to the performance of top U.S. rivals. Moonshot says the model is particularly strong at parametric modeling for mechanical parts and architecture. It can write and run scripts for Blender without special connectors.

Despite the performance boost, the price drops. Token prices fall by 33 percent. Input costs are now $2 per million tokens, down from $3. Output costs drop to $10 from $15. Cache hits cost $0.20. The model retains its size, with 2.8 trillion total parameters and 104 billion active ones. It supports a context window of one million tokens. According to Artificial Analysis, no model with a comparable parameter count scores higher on the Intelligence Index.

Moonshot has also made deliberate choices regarding cybersecurity. For the open-weights release, the company stripped cyber-related data from both training phases. This was done to balance offensive and defensive capabilities. On an internal test that reproduces recent software vulnerabilities, the uncleaned version solved 90.5 percent of tasks. After cleaning, the score dropped to 81.0 percent. This is higher than the previous model but lower than the uncleaned version.

The next step is clear. Moonshot plans to release the model weights by early November. This will make it the top open-weights model on the market. The company believes that while open models should not push the absolute frontier of offensive cyber capabilities, they serve important defensive purposes. As the gap closes, U.S. competitors will likely face pressure to adjust their pricing and capabilities to stay ahead.

For readers in Texas and the South Central states, this shift matters. It signals that high-performance AI is becoming more accessible and cheaper. Businesses in the region can now access advanced tools that rival the best from U.S. labs. The competition is intensifying, and the cost of entry is dropping. This could accelerate the adoption of AI in local industries, from logistics to creative services.