Moonshot AI's Kimi K3.1 Pro Climbs to Fourth on Intelligence Index, Narrowing Gap With Claude Opus 5.5

Moonshot AI, a Chinese artificial intelligence company, has released Kimi K3.1 Pro, a reasoning model that now ranks fourth on the Artificial Analysis Intelligence Index. The model scored 55 points, an 11-point improvement over its predecessor, Kimi K3, and sits just three points behind the top proprietary systems.

K3.1 Pro is a post-trained version of K3 that received substantially more post-training compute. According to Artificial Analysis, only Anthropic's Claude Opus 5.5 (58), Claude Sonnet 5.5 (56) and Muse Wonder Full (56) score higher. Moonshot plans to publish the model weights by November 4, 2026. If it does, K3.1 Pro would become the highest-scoring open-weights model available, reducing the gap between open and proprietary models from 12 points to 3, assuming no other changes in the meantime.

Alongside the main model, Moonshot is offering K3.1 Pro Highspeed, a variant the company says runs two to three times faster. Both versions are available through Kimi.ai, Kimi Work, Kimi Code and the Kimi API.

The largest improvements came in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro reached 1,846 Elo, according to Artificial Analysis, which is 312 points above K3 and just behind Claude Opus 5.5 (1,866) and Muse Wonder Full (1,857). On Terminal-Bench 4.0, Moonshot reports a score of 65.7 percent, ahead of GPT-6 Astra at 58.2 percent and slightly below Claude Opus 5.5 at 66.4 percent. K3 managed only 13 percent on that test.

Moonshot also reports a sharp drop in hallucinations. On its internal Kimi Hallu Bench Hard, claim-level hallucinations fell from 23.4 percent with K3 to 8.3 percent. The company describes the benchmark as a stress test rather than a measure of general error rates, and notes that GPT-6 Astra still leads on it at 5.5 percent. On internal finance benchmarks, the model gained 7.2 points on Financial Modeling and 11.6 points on Financial Research, and Moonshot says it tops all tested models on OfficeQA Pro v2.

The company also highlights 3D modeling and coding. K3.1 Pro works with game engines including Unity, Unreal Engine and Godot, and can write and run scripts for Blender and CAD software without MCP connectors. On the BenchCAD benchmark it scores 92.7 percent with tools, compared with 96.2 percent for Claude Opus 5.5 and 95.9 percent for GPT-6 Astra. Without tools it reaches 72.8 percent, against 73.0 percent for Claude Opus 5.5. Moonshot has also added a "3D Make" feature on Kimi.com for previewing, editing and exporting 3D models.

The architecture is unchanged from K3: 2.8 trillion total parameters, 104 billion of them active, a one-million-token context window, and support for text and images. Artificial Analysis places it on the top of the Pareto frontier, meaning no model of comparable size scores higher on its index. API prices fall by 33 percent, with input at $2 per million tokens, output at $10 and cache hits at $0.20. Because the model generates 14 percent more output tokens, the cost of running each Intelligence Index task is $1.98, nearly the same as K3's $2.00.

Moonshot devotes a section of its technical blog post to cybersecurity. The company says vulnerability research capabilities grew as post-training scaled up, because cyber-related content was part of the general mid-training corpus and related tasks were in the reinforcement learning mix. For the open-weights release, it removed cyber-related data from both training phases. On its internal xdev-bench-hard, which reproduces recently disclosed vulnerabilities in software such as the Linux kernel, Node.js and JavaScriptCore, the uncleaned model solved 90.5 percent of tasks, versus 81.0 percent after cleaning. DeepSeek-V4.1-Flash scored 85.7 percent and Kimi K3 scored 33.3 percent.

Moonshot says no open-weights model should push the absolute frontier of offensive cyber capabilities, but that it wants to raise what open models can do gradually, since they also support defensive work such as auditing legacy systems. Anthropic previously showed that even GLM-5.3, which scores 45 on the Artificial Analysis index, can write exploits at levels close to Claude Mythos.