Kimi K3.1 Pro Launch, Artificial Analysis Ranking

The story so far

Written automatically from the Oesnada stories in this timeline. Every number is checked against them. The numbers in brackets link to those stories.

Timeline

1. 25 October 2026VentureBeat

   Kimi K3.1 Pro Debut Under Scrutiny Over Undisclosed Influencer Promotion Claims

   Moonshot, the Chinese AI startup behind the Kimi models, unveiled Kimi K3.1 Pro on Friday, a 3-trillion-parameter model built by further training its July release, Kimi K3.

2. Also on our radar23 October 2026Kimi K3.1 Pro Ranks 4th on Artificial Analysis Intelligence Index Behind Three@artificialanlys

   Moonshot's new reasoning model, Kimi K3.1 Pro, has launched in fourth place on the Artificial Analysis Intelligence Index with a score of 55, according to the independent AI benchmarking group Artificial Analysis. That is 11 points above Kimi K3 (max setting), which scored 44. Artificial Analysis says the gains come largely from stronger agentic performance, meaning the model's ability to carry out multi-step tasks, and that K3.1 Pro posted the highest score of any model on Terminal-Bench 4.0. The result matters because of where it leaves open-weights models, whose trained parameters can be downloaded and run by others, relative to proprietary systems. Artificial Analysis says K3.1 Pro will become the top-scoring open-weights model once Moonshot publishes its weights, and that the gap between the leading open-weights and proprietary models on the index would then shrink from 12 points to 3. The firm also notes that the model is the same size as Kimi K3, at 2.8 trillion total parameters with 104 billion active at a time, yet scores 11 points higher. It says this places the model at the top of the Pareto frontier, the best available balance of two measures, in at least one of its comparisons. Weights are not yet public, though Artificial Analysis expects Moonshot to release them soon after launch. The model supports a 1 million-token context window, which is the amount of text it can consider at once, and accepts text and image inputs. Pricing is $2 per million input tokens, $10 per million output tokens and $0.20 per million cache-hit tokens, a 33% cut from Kimi K3 on all three rates.

   Read at the source
3. Also on our radar23 October 2026Kimi Unveils K3.1 Pro, 3T-Class Open-Weight Model With 1M-Token ContextKimi

   Kimi has introduced Kimi K3.1 Pro, the first model in its new K3.1 series and, by the company's description, its most capable open-weight model in the 3T class (roughly three trillion parameters), with a context window of one million tokens. A faster variant, Kimi K3.1 Pro Highspeed, runs two to three times as fast for more responsive use. Both are available now on Kimi.ai, Kimi Work, Kimi Code and the Kimi API, and Kimi says the full model weights will be released by Nov 4. The company says K3.1 Pro is a post-trained version of Kimi K3 that used substantially more reinforcement learning compute across hundreds of downstream domains, and that it matches or approaches the strongest proprietary models on coding, agentic and knowledge benchmarks. On BenchCAD, a test of 3D reconstruction through CAD code generation, it scored 92.7% with tools, compared with 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra. Kimi reports a state-of-the-art 64.4 on OfficeQA Pro v2, a test of financial document analysis, and says its claim-level hallucination rate on an internal stress test fell from 23.4% for K3 to 8.3%. It cautions that this figure does not reflect error rates in everyday use. The model works with game engines including Unity, Unreal Engine and Godot, and can write scripts for tools such as Blender without MCP connectors. A new 3D Make feature on kimi.ai and the Kimi app lets users preview, edit and export models, including files ready for 3D printing. Kimi also detailed how it handled cybersecurity risk before the open-weight release. It removed cyber-related data from mid-training and reinforcement learning stages, and says a checkpoint without those steps solved 90.5% of problems on its in-house xdev-bench-hard vulnerability benchmark at pass@3, versus 81.0% for the final model. The company says the model's cyber capability remains behind that of frontier proprietary models that have been public for at least three months, and argues that wider access can help more teams audit their own code. Kimi Enterprise administrators can enable K3.1 Pro for their organizations, though access is off by default at launch.

   Read at the source

Topics in this story

• Anthropic

Methodology and sources

• Atom feed
• Oesnada

Cite this page

Oesnada, "Kimi K3.1 Pro Launch, Artificial Analysis Ranking", https://oesnada.com/storylines/kimi-k3-1-pro-launch-artificial-analysis-ranking, accessed 25 October 2026.