Qwen - AshokBhat/ml GitHub Wiki

About

  • LLM family built by Alibaba Cloud

Models

  • Size: 0.5B, 1.8B, 4B, 7B, 14B, 32B, 72B, MoE-A2.7B
  • Variants: Base model, -Chat, -Chat-AWQ, -Chat-GGUF, -Chat-GPTQ-Int4, -Chat-GPTQ-Int8

Qwen Open-Source LLMs (Alibaba, August 2025)

Model Name Release Date Parameters (B) Specialization / Notes
Qwen3-235B-A22B Apr 2025 235 total / 22 active Flagship MoE model, multilingual, reasoning
Qwen3-30B-A3B Apr 2025 30 total / 3 active MoE, efficient reasoning
Qwen3-32B Apr 2025 32 Dense, general-purpose
Qwen3-14B Apr 2025 14 Dense, balanced performance
Qwen3-8B Apr 2025 8 Dense, efficient for mid-range tasks
Qwen3-4B Apr 2025 4 Dense, lightweight
Qwen3-1.7B Apr 2025 1.7 Dense, edge-capable
Qwen3-0.6B Apr 2025 0.6 Smallest, runs on low-end devices
------------------------ -------------- ------------------------- ---------------------------------------------
Qwen2.5-Coder-32B Nov 2024 32 Code generation, GPT-4o-level coding
Qwen2.5-Coder-14B Nov 2024 14 Code generation
Qwen2.5-Coder-7B Nov 2024 7 Code generation
Qwen2.5-Coder-3B Nov 2024 3 Code generation
Qwen2.5-Coder-0.5B Nov 2024 0.5 Code generation, edge-capable
Qwen2.5-72B 2024 72 General-purpose
Qwen2.5-14B-Instruct 2024 14 Instruction-tuned
Qwen2.5-7B-Instruct 2024 7 Instruction-tuned
------------------------ -------------- ------------------------- ---------------------------------------------
QwQ-32B Nov 2024 32 Reasoning-focused, strong benchmarks

See also