Qwen2.5-Coder
Alibaba's 0.5B–32B open code model family; the 32B rivaled GPT-4o on coding benchmarks.
About
Qwen2.5-Coder (November 2024) is Alibaba's family of code-specialized open-weight models in six sizes from 0.5B to 32B, trained on 5.5 trillion tokens of code and code-adjacent data with 128K context. The 32B-Instruct variant became the reference open code model of its generation, and the smaller sizes made it a default for local coding assistants via Ollama and LM Studio. Everything except the 3B size is Apache 2.0 (the 3B carries a research-only license). Superseded by Qwen3-Coder for frontier work, but still the most practical dense open coder for fine-tuning and local use.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team-reported (technical report): Qwen2.5-Coder-32B-Instruct — HumanEval 92.7, MBPP 90.2, comparable to GPT-4o on coding tasks.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: DeepNewz + HF README + technical report (arXiv 2409.12186) read this session: Nov 11 2024, 0.5B-32B, Apache 2.0 (3B research-only caveat noted), 128K context, 5.5T tokens.; Team-reported: 32B-Instruct HumanEval 92.7, MBPP 90.2, comparable to GPT-4o..