Kimi K3.1 Pro officially unveiled: 3T parameters and a 1M-token context window
Kimi has introduced the first model in its K3.1 family, featuring 3 trillion parameters and a 1-million-token context window, with full weights expected on November 4.
Kimi K3.1 Pro has officially been introduced.
The team released the first model in the K3.1 family, and the results look promising.
- A 3-trillion-parameter model with a 1-million-token context window.
- Outperforms Astra and Fable 5.1 at significantly lower cost.
- Scored 65.7% on Terminal-Bench 4.0.
- Full model weights will be published on November 4.
The model is already available in Kimi, Kimi Work, Kimi Code, and via API.
Kimi is back in the game.
Why it matters
AnalysisIf the claimed specs hold, the model could offer a strong price-performance option for agentic and coding workloads. Publishing full weights would let developers independently evaluate and deploy it.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
Kimi K3.1 Launches as a 2.8T-Parameter AI Model
Kimi K3.1 is described as a very large open model with 2.8 trillion parameters and 104 billion active parameters, but its weights have not yet been released.
Kimi K3.1: teaser hints at new model from the Kimi team
The Kimi team published a teaser suggesting a fast arrival of Kimi K3.1, expected to be the best among available Chinese AI models and significantly cheaper than Kimi K3.
Part I. AI Agent Architecture: The Foundations
The eleven foundational building blocks of modern autonomous agents: model taxonomy, the agentic runtime loop, advanced prompting, distillation, skills architecture, memory systems, function calling, and FinOps.
Part II. AI Agent Architecture: From Model to Memory and Tools
A deep engineering dive: tokenization, KV-cache, sampling determinism, the training stack, quantization, context engineering, the safety stack, production RAG, ReAct, and multi-agent swarm systems.
DeepSeek (Chinese Breakthrough in Open AI)
A series of groundbreaking open models from the Chinese lab DeepSeek (V3, R1). Demonstrates top-tier coding and reasoning capabilities at a cost 10–20 times lower than Western commercial counterparts.
DeepSeek-R1 (DeepSeek Reasoning Model)
A groundbreaking open weights reasoning model based on a 671B MoE architecture, demonstrating the capability for advanced logical reasoning through pure Reinforcement Learning (GRPO).