First posted on Instagram
China's Moonshot AI has launched Kimi K3.1 Pro, a new model the company says scores 65.7% on Terminal-Bench 4.0, a coding-agent benchmark — ahead of OpenAI's GPT-6 Astra at 58.2% and just behind Anthropic's Claude Opus 5.5 at 66.4%.
The launch is Moonshot's first major upgrade since the K3 release. The company describes K3.1 Pro as a post-trained version of Kimi K3, carrying roughly 2.8 trillion total parameters — with 104 billion active at any one time through a mixture-of-experts architecture — and a 1-million-token context window.
Moonshot is shipping the model now through its apps and API, with a faster companion variant called Kimi K3.1 Pro Highspeed, and has committed to publishing the full open weights by November 4, which would make it the strongest open-weight model in the world.
On pricing, Moonshot lists the API at $2 per million input tokens and $10 per million output tokens — about a third cheaper than its predecessor, and potentially more affordable than competing closed models.
The company also says it removed cyber-related training data from the open-weight release to reduce the risk of misuse.
The benchmark figures are Moonshot's own, reported by Unite.AI, Tom's Hardware and others, and have not yet been independently verified. Artificial Analysis reportedly measured the model at 55 on its Intelligence Index, fourth overall. As the company itself concedes, competitors and independent researchers would need to examine the testing methods, real-world performance, safety measures and full costs before drawing firm conclusions about which model is superior.