Moonshot AI's Kimi K3.1 Pro Narrows Gap To Top Models By 3 Points

Chinese AI developer Moonshot AI has released Kimi K3.1 Pro, a model that has closed the performance gap with leading US competitors to just three points. The new system now ranks fourth on the Artificial Analysis Intelligence Index with a score of 55, positioning it directly behind Anthropic’s Claude Opus 5.5, which holds the top spot with 58 points. This release marks a significant shift in the global AI landscape, as the distance between the highest-scoring open-weights model and the best proprietary systems shrinks from a twelve-point lead to a mere three-point difference, assuming the planned release of model weights proceeds as scheduled.

The model represents a major evolution from its predecessor, Kimi K3. Moonshot utilized significantly more post-training compute to refine the system, resulting in an 11-point jump in the overall intelligence score. While Claude Opus 5.5 and Claude Sonnet 5.5 remain ahead, K3.1 Pro now outperforms several other major proprietary offerings. If the company releases the model weights by November 4, 2026, as planned, K3.1 Pro will become the highest-scoring open-weights model available on the market.

Performance improvements are most visible in agentic tasks, where the model shows substantial gains. On the GDPval-AA benchmark, K3.1 Pro achieves 1,846 Elo, a 312-point increase over the previous version. This places it just behind Claude Opus 5.5 and Muse Wonder Full. In coding and terminal tasks, the model scores 65.7 percent on Terminal-Bench 4.0, surpassing GPT-6 Astra and trailing only Claude Opus 5.5. The previous K3 model scored only 13 percent on that same test, highlighting the rapid pace of development.

Reliability has also improved, with a sharp drop in hallucination rates. On Moonshot’s internal Kimi Hallu Bench Hard, claim-level hallucinations fell from 23.4 percent to 8.3 percent. While GPT-6 Astra still leads this specific metric at 5.5 percent, Moonshot notes that its benchmark serves as a stress test rather than a general measure of error rates. The model also shows gains in financial modeling and research, as well as general office tasks, where it reportedly tops all tested models on OfficeQA Pro v2.

The update includes new capabilities for 3D modeling and direct integration with game engines such as Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, the model scores 92.7 percent with tools and 72.8 percent without, closely matching the performance of top proprietary systems. Moonshot has also launched a feature called 3D Make on Kimi.com, allowing users to preview, edit, and export 3D models directly. The company states the model is particularly strong in parametric modeling of mechanical parts and architectural designs.

Despite the performance gains, the model retains the same architectural dimensions as its predecessor, with 2.8 trillion total parameters and 104 billion active ones. However, API prices have dropped by 33 percent. Input costs are now $2 per million tokens, output costs are $10, and cache hits cost $0.20. This pricing structure keeps the cost per Intelligence Index task at $1.98, nearly identical to the previous version, despite the model generating 14 percent more output tokens.

Addressing cybersecurity concerns, Moonshot detailed its approach to offensive capabilities in a technical blog post. The company observed that vulnerability research capabilities grew as post-training scaled, so it stripped cyber-related data from both training phases for the open-weights release. This reduced the model’s score on an internal benchmark from 90.5 percent to 81.0 percent. While this is lower than DeepSeek-V4.1-Flash, it is far higher than the previous K3 model. The company aims to gradually raise the capabilities of open models for defensive purposes while avoiding pushing the absolute frontier of offensive cyber abilities.

Both K3.1 Pro and a faster variant called K3.1 Pro Highspeed are now available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. The Highspeed version runs two to three times faster according to the company. With the model weights expected to be public by early November, the release is set to intensify competition in the open-source AI sector, offering developers a high-performance alternative that competes closely with the industry’s leading proprietary systems.