Moonshot AI released Kimi K3.1 Pro — an MoE model with 2,8 trillion total parameters, 104 billion active parameters and a context of 1 million tokens. MarkTechPost reports this.
In the publication's summary, the model is said to have 65,7% in Terminal-Bench 4.0. The price is stated at $2 per million input tokens and $10 per million output tokens.
Open weights, according to the same message, are expected to appear by 4 November 2026 year. This could expand opportunities for independent evaluation and deployment of the model, but the available package does not contain details on testing methodology, licensing, or actual availability of weights.
editorial commentary
Why it matters
Likely consequence — increased interest in long-context models and open weights. The next observable signal will be the actual release of weights and terms of use. Significant uncertainty arises from the fact that the information is based on metadata from a single publication, without primary confirmation and testing details.