Moonshot AI published the complete weights for Kimi K3.1 Pro on Hugging Face Inc. (Hugging Face) on November 4, making the 2.8-trillion-parameter mixture-of-experts model available for download without a waitlist or gate1. The release spans 1.56 terabytes across 96 safetensors shards. Kimi K3.1 Pro strongly tops the Agent Arena Open-Source Leaderboard, according to reporting from aibase.com2.
Architecture and hardware demands
Only about 104 billion parameters are active on any given token, a design that keeps inference cost well below what the 2.8-trillion headline figure implies. Running the model at usable speed still calls for roughly eight Nvidia Corp. (Nvidia) B300 GPUs and 2.3 terabytes of combined GPU memory. The model eclipses Deep Seek's V4-Pro-Preview at 1.6 trillion parameters and is reported to be the largest Chinese open-weight release yet by total parameter count.
Kimi K3.1 Pro handles multimodal input, including images alongside text, and carries a context window of 1,048,576 tokens. Moonshot AI says the model is tuned for coding, agentic workflows, and financial research.
Benchmarks and pricing
Moonshot AI first put Kimi K3.1 Pro behind an API on October 23. The model scored 55 on the Artificial Analysis intelligence index, three points behind Anthropic's Claude Opus 5.5 at 58. FourWeekMBA's analysis says Kimi K3.1 Pro leads Claude Opus 5.5 on 16 of 37 individual benchmark tests. Moonshot AI has been selling Kimi K3.1 Pro through its API at roughly half what Claude Opus 5.5 costs.
License terms and safety posture
The Kimi K3.1-Pro License permits use, modification, fine-tuning, and redistribution. Companies offering the model as a hosted API service with more than $20 million in trailing twelve-month revenue must sign a separate commercial agreement with Moonshot AI.
On the safety side, Moonshot AI says Kimi K3.1 Pro's offensive security skills stay behind proprietary frontier models that have already been public for at least three months. The company has said it aims not to release open weights that extend the absolute frontier of cyber capability.
ANALYSIS The revenue-gated license creates a two-tier access structure: smaller developers and researchers get unrestricted use, while large-scale API hosts face a negotiation requirement that preserves Moonshot AI's commercial leverage over the highest-volume deployments.