Tech & AI
China's Moonshot launches Kimi K3.1 Pro at lower API prices and plans to release its full weights by November 4
The Beijing-based lab says its roughly 3-trillion-parameter model nearly matches the best US systems on a key coding test, while it stripped hacking material from training before the open release.
HedgeWeekly Tech & AI Desk ·
Chinese AI developer Moonshot AI released Kimi K3.1 Pro on Friday, an upgraded version of its Kimi K3 model, and said it plans to publish the full model weights by November 4. Open weights are the trained parameters of a model; releasing them lets anyone download, run and modify the system on their own hardware. The model is already available through Kimi's apps and its paid API, alongside a Highspeed variant that Moonshot says runs two to three times faster.
According to technical details reported by MarkTechPost, K3.1 Pro is a mixture-of-experts model, a design that switches on only part of the network for each request, with about 2.8 trillion parameters in total and 104 billion active at a time. It handles a context window of 1 million tokens, roughly the amount of text the model can consider at once. API pricing is $2 per million input tokens and $10 per million output tokens, below the $3 and $15 charged for the original K3, with a Highspeed tier at $4 and $20.
On Terminal-Bench 4.0, a test of AI coding agents, Moonshot reported a score of 65.7%, against 66.4% for Anthropic's Claude Opus 5.5 and 58.2% for OpenAI's GPT-6 Astra; the figures are the company's own and have not been independently verified. Moonshot also said it removed cyberattack-related data from training before the open release: on its internal exploit test, an unfiltered version solved 90.5% of tasks, while the shipped model solved 81%.
Why it matters: each near-frontier open-weight release from China narrows the price premium US labs can charge for top-tier models, a pressure point for the revenue assumptions behind private valuations at OpenAI and Anthropic and for the cloud providers reselling their models. The cyber-data filtering is also likely to feed the policy debate in Washington over open models and export controls. Investors should watch independent benchmark results once the weights are published on November 4, and how quickly cloud platforms and enterprises adopt the model.