Kimi K3.1 Pro Closes In on Claude, Just 3 Points Behind
Moonshot AI launched the Kimi K3.1 Pro model with 2.8 trillion parameters, ranking fourth on the Artificial Analysis index with 55 points, just 3 points behind Claude Opus 5.5.
Chinese company Moonshot AI launched the "Kimi K3.1 Pro" model, a post-trained version of "Kimi K3" with additional compute, ranking fourth on the "Artificial Analysis" intelligence index with 55 points. This represents an 11-point leap from the previous version, bringing the model within just three points of "Claude Opus 5.5," which leads with 58 points.
The model retains its previous size at 2.8 trillion total parameters and 104 billion active parameters, with a context window of 1 million tokens, and the capability to process text and images. Moonshot plans to release the model weights by November 4, which could make it the highest-scoring open-weights model and narrow the gap between open and closed models from 12 points to 3 points.
The largest improvements are focused on agentic tasks and coding, as the model scored 65.7% on the "TerminalBench 4.0" test, compared to 58.2% for "GPT-6 Astra" and 66.4% for "Claude Opus 5.5." The hallucination rate in an internal test dropped from 23.4% to 8.3%, according to the company. Additionally, API pricing was reduced by 33%, bringing input pricing down to two dollars per million tokens instead of three.
Moonshot stated that it removed cybersecurity data from the training stages prior to releasing the open-weights version, as it does not want an open model to push the boundaries of offensive capabilities. It also announced a "High Speed" version that operates two to three times faster. The models are available via Kimi platforms and the API.
What do these terms mean?
Parameter: An internal unit in a neural network tuned during training, its count is used as an approximate measure of model size.
Context Window: The maximum amount of text and images the model can consider in a single prompt, usually measured in tokens.
Open Weights: Model files published so anyone can download, run, and modify them, rather than keeping them restricted to the company.
Hallucination: When a model presents information that sounds convincing but is incorrect or unsupported by the source.
Weekly Newsletter
Read between the lines before everyone else. Decode the most important economic, tech, and decision-maker movements in the region.. in 5 minutes every Saturday.
