Moonshot AI's New Model Narrows the Gap With Top Proprietary Rivals

The competitive landscape for artificial intelligence just shifted as a Chinese model surged into the top tier of global rankings.

Moonshot AI has introduced Kimi K3.1 Pro, a post-trained iteration of its previous model that utilizes significantly more compute during the final training phase. This update places the system at fourth on the Artificial Analysis Intelligence Index with a score of 55. It trails only Claude Opus 5.5, which holds 58 points, and Claude Sonnet 5.5 and Muse Wonder Full, which both sit at 56. This represents an eleven-point improvement over the prior version. If the company releases the model weights by November 4, 2026, as planned, this would become the highest-scoring open-weights option available. Such a move would reduce the performance gap between open and proprietary systems from twelve points to just three, assuming no other major releases occur in the interim.

The most notable gains appear in agentic tasks, where the model demonstrates substantial progress. On the GDPval-AA benchmark, the new system reaches 1,846 Elo, a jump of 312 points from its predecessor. This places it just behind Claude Opus 5.5 and Muse Wonder Full. In coding evaluations, specifically Terminal-Bench 4.0, Moonshot reports a score of 65.7 percent. This figure exceeds GPT-6 Astra and sits very close to Claude Opus 5.5. The previous model managed only 13 percent on this specific test. Additionally, internal testing shows a sharp reduction in hallucinations. Claim-level errors dropped from 23.4 percent to 8.3 percent on a difficult internal benchmark, though GPT-6 Astra still leads with a 5.5 percent error rate.

Beyond general reasoning, the model shows strength in specialized fields like 3D modeling and finance. It can interact directly with game engines such as Unity and Unreal Engine. On the BenchCAD benchmark, it scores 92.7 percent when using tools, which is close to top proprietary rivals. Without external tools, it achieves 72.8 percent, nearly matching Claude Opus 5.5. The company states it is particularly strong in parametric modeling for mechanical and architectural parts. Users can now preview and export 3D models through a new feature on the company's website. Financial benchmarks also saw improvements, with gains in both financial modeling and research tasks.

A key differentiator for this release is cost. While the parameter count remains at 2.8 trillion total with 104 billion active, API prices have dropped by 33 percent. Input costs are now $2 per million tokens, down from $3. Output costs fall to $10 from $15. Cache hit costs decrease to $0.20. This pricing strategy positions the model as a high-value option for developers who need robust performance without the premium price tag of the leading proprietary systems. The model supports a one-million-token context window and handles both text and images.

Moonshot has also addressed cybersecurity concerns by modifying how the model is trained. The company noted that vulnerability research capabilities grew as post-training scaled. To prepare for an open-weights release, they removed cyber-related data from both training phases. This adjustment reduced the model's ability to solve specific security tasks from 90.5 percent to 81.0 percent on an internal benchmark. While this is still high, the company argues it strikes a balance between offensive capability and defensive utility, such as auditing legacy systems. They aim to gradually raise the ceiling for open models while maintaining a deliberate limit on pure offensive power.

The new model is available now through various platforms, including a high-speed variant that runs two to three times faster. For California businesses and tech firms, this release signals that the race between open and closed models is tightening. With prices dropping and performance climbing, enterprises may soon have more leverage when negotiating AI contracts. The coming weeks will be critical as the industry waits for the open-weights release in early November. That event will determine if the three-point gap remains or if other vendors respond with their own updates.