Moonshot launched Kimi K3.1 Pro on October 26, touting a one-million-token context window and improved coding and agent capabilities.

Moonshot AI introduced Kimi K3.1 Pro on October 26, positioning it as the first model in a new series and its most capable open-weight model in the 3-trillion-parameter class. The model retains Kimi K3’s reported mixture-of-experts design—2.8 trillion total parameters, about 104 billion active—and supports a context window of up to 1,048,576 tokens. Moonshot says post-training improvements target coding, financial research, 3D creation and long-running agent tasks; on its internal Kimi Code Bench 3.0 long-horizon test, the score rose to 61.2 from K3’s 55.95. API pricing fell to $2 per million uncached input tokens and $10 per million output tokens, from K3’s $3 and $15. Artificial Analysis gave K3.1 Pro an Intelligence Index score of 55, fourth in its comparison, and estimated $2.11 per index task. The upgrade builds on K3, launched in July with a million-token context, and intensifies competition with US frontier models. Moonshot says it will publish K3.