80B total / 3B active (MoE)
Apache 2.0
Open-weight coding-focused model designed for coding agents and local development with high efficiency; model card states performance comparable to models with 10-20x more active parameters while activating only 3B out of 80B total. Architecture is a 48-layer hybrid stack following 12 × (3 × (Gated DeltaNet -> MoE) -> 1 × (Gated Attention -> MoE)); Gated DeltaNet uses 32 linear attention heads for V and 16 for QK (head dim 128), while Gated Attention uses 16 Q heads and 2 KV heads (head dim 256, RoPE dim 64). MoE configuration uses 512 experts with 10 activated experts plus 1 shared expert per token and expert intermediate dim 512. Native context length is 262,144 tokens for repository-scale code understanding. Release notes emphasize advanced agentic behavior for long-horizon reasoning, complex tool usage, and recovery from execution failures, plus integration across developer tooling and scaffolds (including CLI/IDE environments such as Claude Code, Qwen Code, Qoder, Kilo, Trae, and Cline). The model is non-thinking mode only and does not generate <think></think> blocks. Deployment guidance highlights support in modern serving stacks including SGLang and vLLM, with additional ecosystem support referenced for other runtimes.