Moonshot AI Narrows Gap With Anthropic via Kimi K3.1 Pro

A 33 percent reduction in API costs marks the most immediate financial shift for developers as Moonshot AI rolls out Kimi K3.1 Pro, a significant upgrade to its previous reasoning model. This post-trained variant, which incorporates substantially more compute during the training phase, has surged to fourth place on the Artificial Analysis Intelligence Index with a score of 55 points. This performance places it just three points behind the leading proprietary models, signaling a tightening race in the global AI landscape.

The model currently trails only Claude Opus 5.5, which holds 58 points, and is tied for third with Claude Sonnet 5.5 and Muse Wonder Full, both at 56. Should Moonshot release the model weights as scheduled for November 4, 2026, Kimi K3.1 Pro would become the highest-scoring open-weights model available. This move would narrow the performance gap between open and proprietary systems from 12 points down to a mere three, assuming no other major releases occur in the interim. A faster variant, K3.1 Pro Highspeed, which operates two to three times quicker, is also available immediately across Kimi.ai, Kimi Work, Kimi Code, and the API.

Performance gains are most visible in agentic tasks. On the GDPval-AA benchmark, the new model achieves an Elo rating of 1,846, a 312-point increase over its predecessor. It now sits slightly behind Claude Opus 5.5 and Muse Wonder Full. In terminal-based coding tests, specifically Terminal-Bench 4.0, Moonshot reports a score of 65.7 percent, surpassing GPT-6 Astra and coming within half a percent of Claude Opus 5.5. Notably, the previous K3 model managed only 13 percent on that same evaluation, highlighting a dramatic leap in capability.

Reliability has also improved, with claim-level hallucinations dropping from 23.4 percent to 8.3 percent on Moonshot’s internal Kimi Hallu Bench Hard. While GPT-6 Astra maintains a lead with a 5.5 percent rate, Moonshot notes this is a stress test rather than a general error measure. The model also outperforms all tested competitors on OfficeQA Pro v2 for general office tasks, while gaining points in financial modeling and research benchmarks.

In the realm of 3D modeling, Kimi K3.1 Pro integrates directly with game engines such as Unity, Unreal Engine, and Godot. It scores 92.7 percent on the BenchCAD benchmark when using tools, approaching the performance of Claude Opus 5.5 and GPT-6 Astra. Without tools, its score of 72.8 percent nearly matches the top proprietary model. The company has also launched a "3D Make" feature on its website, allowing users to preview, edit, and export 3D models, with particular strength in parametric modeling for mechanical and architectural applications.

The model retains the same architecture as its predecessor, featuring 2.8 trillion total parameters and 104 billion active ones, with a one-million-token context window. Despite the identical size, the cost per Intelligence Index task remains nearly flat at $1.98, as the model generates 14 percent more output tokens. For the upcoming open-weights release, Moonshot has deliberately stripped cybersecurity data from its training phases. This reduces the model’s ability to solve offensive cyber tasks from 90.5 percent to 81.0 percent on internal benchmarks, a calculated limit to balance offensive and defensive capabilities.