Kimi K2.6

Kimi K2.6 is an open, native multimodal agentic model created by Moonshot AI, released in April 2026. It builds on the architecture and approach established by Kimi K2.5, extending capabilities in long-horizon coding, coding-driven design, proactive autonomous execution, and swarm-based task orchestration.

Key Features

• Long-Horizon Coding — significant improvements on complex, end-to-end coding tasks across Rust, Go, Python and domains including front-end, DevOps and performance optimization.
• Coding-Driven Design — transforms simple prompts and visual inputs into production-ready interfaces and lightweight full-stack workflows.
• Elevated Agent Swarm — horizontal scaling to 300 sub-agents executing 4,000 coordinated steps, dynamically decomposing tasks into parallel, domain-specialized subtasks.
• Proactive & Open Orchestration — supports persistent, 24/7 background agents that proactively manage schedules, execute code and orchestrate cross-platform operations.

Architecture

Kimi K2.6 is a Mixture-of-Experts model with 1T total parameters and 32B active parameters across 61 layers (1 dense), 384 experts per MoE layer, 8 experts selected per token, and 1 shared expert. The attention mechanism is MLA (Multi-head Latent Attention) with 64 heads and 7168 hidden dimension. Activation: SwiGLU. Vocabulary: 160K. Context: 256K tokens. Vision encoder: MoonViT (400M parameters). The model uses native INT4 quantization (same as Kimi K2 Thinking).

Modes and Access

The model supports Thinking mode (recommended temperature 1.0) and Instant mode (recommended temperature 0.6), as well as preserve_thinking (retains reasoning across turns in coding-agent scenarios). Weights are released under the Modified MIT License. The API is available at platform.moonshot.ai (OpenAI and Anthropic compatible). Recommended inference engines: vLLM, SGLang, KTransformers. Dedicated coding-agent framework: Kimi Code CLI (kimi.com/code).