{
 "slug": "Qwen3-Coder-Next",
 "title": "Qwen3-Coder-Next",
 "type": "model",
 "short_desc": "Alibaba Qwen team's February 2026 open-weight coding model, an ultra-sparse 80B hybrid-attention MoE tuned for coding agents and local development.",
 "categories": [
  "Alibaba models",
  "2026 model releases",
  "Open-weight models",
  "Mixture-of-experts models",
  "Models"
 ],
 "infobox": {
  "Developer": "Alibaba (Qwen team)",
  "Family": "Qwen",
  "Announced": "February 2026",
  "Architecture": "Hybrid linear-attention design (Gated DeltaNet plus gated attention) with ultra-sparse mixture-of-experts layers",
  "Parameters": "80B total, roughly 3B activated per token (developer-reported)",
  "Context window": "256,000 tokens native (developer-reported)",
  "Post-training": "Large-scale executable task synthesis, environment interaction, and reinforcement learning for agentic coding",
  "License / access": "Apache 2.0 open weights; downloadable via Hugging Face and served through Alibaba Cloud",
  "Paper/report": "Qwen team, \"Qwen3-Coder-Next: Pushing Small Hybrid Models\" (2026)"
 },
 "infobox_links": {
  "Developer": [
   "Qwen_team"
  ],
  "Family": [
   "Qwen_(model_family)"
  ],
  "Architecture": [
   "Mixture_of_experts"
  ],
  "License / access": [
   "Hugging_Face"
  ],
  "Paper/report": [
   "Qwen_team"
  ]
 },
 "aliases": [
  "Qwen 3.6"
 ],
 "family": "Qwen (model family)",
 "org": "Alibaba",
 "registry_status": "stub",
 "words": 381,
 "references": 3,
 "outbound": [
  "Claude_Opus_4.5",
  "Code_Llama",
  "Codestral",
  "DeepSeek-Coder-V2",
  "Devstral",
  "GLM-4.6",
  "GPT-5.1",
  "Hugging_Face",
  "Instruction_tuning",
  "Kimi_K2",
  "Large_language_model",
  "Mixture_of_experts",
  "Qwen3",
  "Qwen3-Coder",
  "Qwen_(model_family)",
  "Qwen_team",
  "Transformer_(architecture)"
 ],
 "inbound": [
  "Qwen3-Coder",
  "Qwen_(model_family)"
 ],
 "url": "wiki/Qwen3-Coder-Next.html",
 "built": "2026-07-24"
}