Chinese AI Firm Releases Kimi K3.1 Pro, Closing Gap With Top Rivals

Developers using artificial intelligence models for coding and data analysis now have a new option from China. Moonshot AI has released Kimi K3.1 Pro, a reasoning model that ranks fourth on the Artificial Analysis Intelligence Index. The model achieved a score of 55 points. This represents an 11 point improvement over its predecessor, Kimi K3. It currently trails the top proprietary systems by only three points.

Only three models currently score higher on the index. Anthropic's Claude Opus 5.5 leads with 58 points. Claude Sonnet 5.5 and Muse Wonder Full both score 56 points. Moonshot plans to publish the model weights by November 4, 2026. If the release goes as planned, K3.1 Pro will become the highest-scoring open-weights model available. This move would reduce the gap between open and proprietary models from 12 points to 3 points, assuming no other changes occur.

The company offers two versions of the new model. The standard K3.1 Pro focuses on accuracy. The K3.1 Pro Highspeed variant runs two to three times faster. Both are available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. API prices drop by 33 percent with this release. Input costs sit at $2 per million tokens, while output costs are $10 per million tokens. Cache hits cost $0.20 per million tokens.

Performance gains are most visible in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro reached 1,846 Elo. This is 312 points above the previous version. It sits just behind Claude Opus 5.5, which scored 1,866 Elo, and Muse Wonder Full, which scored 1,857 Elo. On Terminal-Bench 4.0, the model scored 65.7 percent. This beats GPT-6 Astra, which scored 58.2 percent, but falls slightly short of Claude Opus 5.5, which scored 66.4 percent. The older Kimi K3 managed only 13 percent on that same test.

Moonshot reports a sharp decrease in hallucinations. On its internal Kimi Hallu Bench Hard, claim-level hallucinations dropped from 23.4 percent with K3 to 8.3 percent. The company describes this benchmark as a stress test rather than a measure of general error rates. GPT-6 Astra still leads this specific metric with a 5.5 percent hallucination rate. Internal finance benchmarks show gains of 7.2 points in Financial Modeling and 11.6 points in Financial Research.

The model maintains the same architecture as Kimi K3. It features 2.8 trillion total parameters, with 104 billion active. The context window supports one million tokens and handles both text and images. Artificial Analysis places it at the top of the Pareto frontier for its size. The architecture change in post-training compute drove the performance improvements. Moonshot added a 3D Make feature on Kimi.com for previewing and exporting 3D models. The model now works directly with Unity, Unreal Engine, and Godot without special connectors.

Security experts note a shift in how Moonshot handles cybersecurity capabilities. For the open-weights release, the company removed cyber-related data from both training phases. This reduction aims to balance offensive and defensive utility. On an internal test reproducing recent software vulnerabilities, the uncleaned model solved 90.5 percent of tasks. After removing that data, the score fell to 81.0 percent. DeepSeek-V4.1-Flash scored 85.7 percent on the same test, while the older Kimi K3 scored 33.3 percent. Moonshot states it wants to raise open model capabilities gradually to support defensive work like auditing legacy systems.