Moonshot AI unveils kimi K3.1 pro: a 3t-class model with 1m context window and open weights coming in november 2026

Moonshot AI has launched Kimi K3.1 Pro, a 3T-class open-weight model with a 1-million-token context window. The model is positioned to compete with top proprietary models in coding, agentic, and knowledge benchmarks. It is available via API with pricing of $2.00 input and $10.00 output per million tokens, with cache hits at $0.20. A high-speed variant, Kimi K3.1 Pro Highspeed, runs 2 to 3 times faster at $4.00 input and $20.00 output. Open weights are set to release in November 2026. The model is based on the K3 architecture, which uses Kimi Delta Attention and Attention Residuals, activating 16 of 896 experts through a Stable LatentMoE framework. Moonshot claims the model solves 81.0% on its internal xdev-bench-hard test, though it lags behind proprietary models in offensive cyber tasks. The model is accessible through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API, with endpoints compatible with OpenAI. Moonshot removed cyber-related data from mid-training and reinforcement learning for the open-weight release. The model is live with API IDs kimi-k3.1-pro and kimi-k3.1-pro-highspeed. Pricing for K3.1 Pro is lower than K3’s $3.00 input and $15.00 output rates.