Kimi K2.6
Moonshot's open-weight 1T MoE flagship for long-horizon coding and multi-agent orchestration with 300-subagent Agent Swarm system
Kimi K2.6
Moonshot AI • April 2026
Training Data
Up to early 2026
Kimi K2.6
April 2026
Parameters
1 trillion (32B active)
Training Method
Sparse MoE (384 experts, 8 routed + 1 shared), MuonClip optimizer, SwiGLU
Context Window
262,144 tokens
Knowledge Cutoff
Not disclosed
Key Features
Agent Swarm (300 sub-agents) • Multi-turn Tool Calling • Vision Inputs • Open Weights (Modified MIT)
Capabilities
Coding: Outstanding
Agentic: Outstanding
Vision: Excellent
What's New in This Version
Same 256K context as K2.5 but with greatly improved long-context stability; new Agent Swarm scales to 300 domain-specialized sub-agents and 4,000 coordinated steps per autonomous run
Moonshot's open-weight 1T MoE flagship for long-horizon coding and multi-agent orchestration with 300-subagent Agent Swarm system
What's New in This Version
Same 256K context as K2.5 but with greatly improved long-context stability; new Agent Swarm scales to 300 domain-specialized sub-agents and 4,000 coordinated steps per autonomous run
Technical Specifications
Key Features
Capabilities
Other Moonshot AI Models
Explore more models from Moonshot AI
Kimi K2.5
Moonshot AI's open-source visual agentic model with native multimodal capabilities and multi-agent orchestration
Kimi K2
Moonshot AI's flagship model - #1 open-source LLM on LMSYS Arena with exceptional agentic capabilities
Kimi K1.5
Moonshot AI's advanced reasoning model matching OpenAI o1 performance
