Moonshot AI Releases Kimi K3.1 Pro: A 2.8T Total/104B Active Params (MoE) Model with 1M Context
What it does
Moonshot AI has launched the Kimi K3.1 Pro, a mixture of experts (MoE) language model with a total of 2.8 trillion parameters and 104 billion active parameters during inference. It supports a massive 1 million token context window, enabling long-term input processing far beyond typical models. The model currently delivers a 65.7 percent score on the Terminal-Bench 4.0 benchmark and will cost $2 per million prompt tokens, $10 per million output tokens. Open weight releases are planned for November 4, 2026.
Why it matters
The Kimi K3.1 Pro pushes parameter scale and context size significantly. The large active parameter count combined with a 1 million token context window could unlock services requiring deep, extensive memory such as complex document analysis, long-term conversations, and detailed multi-step automation. The lower usage costs moonshot AI offers may pressure competitors charging more for similar capabilities. Open weights arriving in 2026 will provide the research and builder community a valuable tool to experiment with extreme-scale MoE models and long context lengths.
Who it is for
Companies building AI products that demand longer term memory or need to handle extensive inputs will find Kimi K3.1 Pro attractive, especially where cost per token is critical. Research teams can explore scaling laws and model efficiencies at unprecedented levels with the open weights release. Investors and vendors watching AI infrastructure may want to track performance improvements in large MoE models handling very long contexts, which could shift cloud cost economics.
The catch
The Kimi K3.1 Pro is a complex MoE model that may require specialized infrastructure to run efficiently, raising implementation and operational challenges. The release timeline delays open weights and full accessibility until late 2026, limiting immediate adoption outside Moonshot AI’s platform. Real-world performance beyond benchmarks is yet to be proven, so early users should temper expectations until practical workloads validate claims.
What to watch next
Monitor how Moonshot AI’s pricing influences market rates for high-end LLM services, especially those offering long context windows. Look for ecosystem developments around supporting ultra-large MoE architectures in cloud environments. The open weight release scheduled for November 2026 will be a critical event to see community adoption, third-party benchmarking, and integrations that confirm the technical merits and practical advantages of the Kimi K3.1 Pro.
AI Quick Briefs Editorial Desk