AI Costs Drop: New Chinese Model Challenges Top Rivals
A new AI model from Moonshot AI lowers costs and boosts performance, narrowing the gap with leading proprietary systems.
By Kwame Silva ·
Developers and businesses using AI tools can now expect lower prices and faster performance. A new model from a Chinese company has cut its API costs by a third. It also runs significantly faster than its predecessor. These changes make advanced AI more accessible for everyday work and coding tasks.
The model, named Kimi K3.1 Pro, currently scores 55 points on the Artificial Analysis Intelligence Index. This places it in fourth position globally. It trails the top models, such as Claude Opus 5.5, by only three points. The previous version, Kimi K3, scored 44 points. The new version represents an 11-point improvement. If the company releases the model's underlying code by November 4, it will become the highest-scoring open-weights model available.
Moonshot AI, the developer, is also launching a high-speed variant. This version runs two to three times faster than the standard model. Both versions are available now through various platforms and APIs. The standard model retains the same size as its predecessor, with 2.8 trillion total parameters. However, the price per million input tokens drops from $3 to $2. Output costs fall from $15 to $10 per million tokens.
Performance gains are most visible in agentic tasks, where the model performs complex, multi-step operations. On a specific benchmark for these tasks, the new model outperforms several competitors. It also shows a dramatic reduction in hallucinations. In internal tests, the rate of incorrect claims dropped from 23.4 percent to 8.3 percent. The company notes that this is a stress test, not a measure of general error rates.
The model also excels in 3D modeling and coding. It can work directly with game engines like Unity and Unreal. On a benchmark for computer-aided design, it scores 92.7 percent when using tools. This is close to the top proprietary models. The company has launched a feature that allows users to preview and export 3D models directly through its website.
The stakes for the AI industry are high. If the open-weights release proceeds as planned, the gap between open and proprietary models will shrink significantly. Currently, the difference is 12 points. With this new release, it could drop to just three points. This shift suggests that open models are becoming more competitive. It may pressure proprietary developers to lower their prices and improve their efficiency.
Moonshot AI has also addressed cybersecurity concerns. The company removed cyber-related data from the training process for the open-weights version. This was done to limit the model's offensive capabilities. On a difficult internal test, the uncleaned version solved 90.5 percent of tasks. After cleaning, the new model solved 81.0 percent. This approach aims to balance capability with safety for the broader user base.
