Moonshot AI's Kimi K3.1 Pro Challenges Leading US Models With Lower Costs

The landscape of artificial intelligence is shifting as a major player from China releases a new model that performs at levels previously reserved for top American competitors. This development marks a notable step forward for open-weights models in the global race for computing power and capability.

Moonshot AI, a Chinese technology company, has officially launched Kimi K3.1 Pro. This version is a post-trained update of its predecessor, Kimi K3, designed with significantly more post-training compute. On the Artificial Analysis Intelligence Index, the new model achieves a score of 55 points. This represents an 11-point improvement over the previous version and places it in fourth position among all available models.

Currently, only three models rank higher than Kimi K3.1 Pro. Anthropic’s Claude Opus 5.5 leads with a score of 58, followed by Claude Sonnet 5.5 and Muse Wonder Full, both of which score 56. Moonshot plans to release the full model weights by November 4, 2026. If this schedule holds, Kimi K3.1 Pro would become the highest-scoring open-weights model on the market. This move would reduce the performance gap between open and proprietary systems from 12 points to just 3 points.

The company also introduced a variant named K3.1 Pro Highspeed. This version operates two to three times faster than the standard release. Both models are immediately available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. Users can access these tools via standard web interfaces or developer platforms.

Performance improvements are most evident in agentic tasks, where the model shows substantial gains. On the GDPval-AA benchmark, Kimi K3.1 Pro 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 specific category. In terminal-based coding tests, the new model scores 65.7 percent on Terminal-Bench 4.0. This result exceeds GPT-6 Astra, which recorded 58.2 percent, and closely matches Claude Opus 5.5 at 66.4 percent.

The model also demonstrates better reliability in factual accuracy. Moonshot reports a sharp decline in hallucination rates. On its internal stress test, claim-level hallucinations dropped from 23.4 percent in the previous version to 8.3 percent. While GPT-6 Astra maintains a lower rate of 5.5 percent, Moonshot notes that its internal benchmark serves as a rigorous test rather than a measure of general error. In financial applications, the new model improves significantly in modeling and research tasks, topping all tested models on the OfficeQA Pro v2 benchmark for general office duties.

Capabilities in 3D modeling and coding have expanded as well. The model can interact directly with game engines such as Unity, Unreal Engine, and Godot. It handles everything from minor fixes to full project builds. On the BenchCAD test, it scores 92.7 percent when using external tools, a result close to leading competitors. Without tools, it achieves 72.8 percent, nearly matching the performance of Claude Opus 5.5. The company has also launched a feature on its website that allows users to preview, edit, and export 3D models directly.

Despite the performance boost, the model retains the same architectural dimensions as its predecessor. It features 2.8 trillion total parameters, with 104 billion active ones. The context window supports one million tokens and processes both text and images. Pricing has dropped by 33 percent. Input costs are now $2 per million tokens, down from $3. Output costs are $10, reduced from $15. Cache hits cost $0.20. These price reductions make the model more accessible for developers and businesses looking to integrate advanced AI capabilities.

Moonshot also addressed cybersecurity considerations in its technical documentation. The company observed that vulnerability research capabilities grew as post-training scaled. To prepare for the open-weights release, they removed cyber-related data from the training process. On an internal benchmark for reproducing known vulnerabilities, the uncleaned version solved 90.5 percent of tasks. After data cleaning, the final model solves 81.0 percent. This is lower than DeepSeek-V4.1-Flash, which scores 85.7 percent, but significantly higher than the previous Kimi K3 version. The company aims to balance offensive capabilities with defensive utility, allowing open models to assist in auditing legacy systems without pushing the absolute limits of cyber offense.

This release underscores the rapid pace of innovation in the AI sector. By combining high performance with lower costs and a commitment to open weights, Moonshot AI is challenging the dominance of established Western providers. As the November deadline for the open-weights release approaches, the industry will watch closely to see if this model reshapes the balance between proprietary and open-source AI systems.