Moonshot AI Releases Kimi K3.1 Pro to Challenge Anthropic's Lead

The race for the most capable artificial intelligence model just got tighter. Moonshot AI, a Chinese developer, has released a new reasoning model called Kimi K3.1 Pro. This update lands in fourth place on the Artificial Analysis Intelligence Index with a score of 55. It sits just three points behind the current leaders, which are Anthropic's Claude Opus 5.5 and two other top proprietary models. For a New England reader, this matters because it suggests the gap between the best closed-source systems and the best open options is shrinking fast.

This is not a sudden jump. Moonshot has been building on the Kimi K3 line. The new K3.1 Pro is a post-trained version of its predecessor. The company used significantly more compute for the post-training phase. That extra work resulted in an 11-point improvement over the original K3. The model is available now through Kimi.ai, Kimi Work, Kimi Code, and the API. There is also a variant called K3.1 Pro Highspeed that runs two to three times faster.

The biggest gains appear in agentic tasks, which involve an AI executing multi-step goals. On the GDPval-AA benchmark, K3.1 Pro reaches 1,846 Elo. That is a 312-point leap from the previous version. It places the model just behind Claude Opus 5.5 and Muse Wonder Full. Coding performance also improved dramatically. On Terminal-Bench 4.0, the new model scores 65.7 percent. This beats GPT-6 Astra and is barely behind Claude Opus 5.5. The older K3 model managed only 13 percent on that same test.

Accuracy has improved as well. Moonshot reports a sharp drop in hallucinations. On its internal stress test, claim-level errors fell from 23.4 percent to 8.3 percent. The company notes this is a hard test and not a measure of general error rates. GPT-6 Astra still has a slight edge at 5.5 percent. In finance, the new model gained points in modeling and research. Moonshot claims it tops all tested models on general office tasks in their OfficeQA Pro v2 benchmark.

One of the standout features is its ability to work with 3D modeling and game engines. The model can handle quick fixes or full project builds in Unity, Unreal Engine, and Godot. It scores 92.7 percent on the BenchCAD benchmark when using tools. Without tools, it hits 72.8 percent, which nearly matches Claude Opus 5.5. Moonshot says the model is strong at parametric modeling for mechanical parts and architecture. It can write and run scripts for Blender and CAD software without special connectors. A new feature called 3D Make on Kimi.com lets users preview and export these models.

Pricing and size remain key factors for developers. K3.1 Pro keeps the same architecture as K3. It has 2.8 trillion total parameters and 104 billion active ones. The context window is one million tokens. Despite the higher performance, API prices drop by 33 percent. Input costs are now $2 per million tokens, down from $3. Output costs are $10, down from $15. Cache hits cost $0.20. The cost per Intelligence Index task is $1.98, which is almost the same as before because the model generates 14 percent more output tokens.

Moonshot also addressed cybersecurity capabilities in its technical release. The company noted that vulnerability research skills grew as post-training scaled up. For the open-weights release, they removed cyber-related data from both training phases. This deliberate limit reduces the model's offensive capabilities. On an internal test of known vulnerabilities, the uncleaned version solved 90.5 percent of tasks. After cleaning, K3.1 Pro solves 81.0 percent. DeepSeek-V4.1-Flash scores 85.7 percent on the same test. The previous Kimi K3 scored 33.3 percent.

The company believes no open-weights model should push the absolute frontier of offensive cyber capabilities. However, they want to gradually raise what open models can do for defensive purposes, such as auditing legacy systems. Anthropic previously showed that even lower-scoring models can write exploits at high levels. Moonshot plans to release the model weights by November 4, 2026. If that happens, K3.1 Pro would become the highest-scoring open-weights model available. That move would shrink the gap between open and proprietary models from 12 points to just three, assuming no other changes occur by then.

The next step is clear. Moonshot is betting that open models can compete with the best proprietary systems while remaining affordable. The release of weights in early November will be the real test. If the performance holds up in the open community, the landscape for AI development could shift significantly. For now, Kimi K3.1 Pro sits at the top of the Pareto frontier for its size. No model with a comparable parameter count scores higher on the Intelligence Index.