Qwen2.5 Coder 32B
The self-hosted coding sweet spot: 32B, 128K, Apache-2.0.
About
Qwen2.5-Coder-32B-Instruct is the coding specialist of the Qwen2.5 family: 32B parameters, 128K-token context, strong HumanEval/MBPP/SWE-bench-adjacent scores — and notably Apache-2.0 licensed (genuinely open source, unlike its Qwen-licensed 72B sibling). 794k Hugging Face downloads, not gated. It became the default self-hosted coding model for 2025: big enough to be capable, small enough for 2x24GB GPUs in Q8. Superseded by Qwen3-Coder, but the Apache-2.0 license keeps it the safest commercial bet in its class.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported (Alibaba, Nov 2024): strong HumanEval/MBPP, competitive with GPT-4o-mini-class on coding — provider-reported. Widely re-benchmarked; treat as historical.
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: HF downloads 793,989 (peer-normalized); released 706 days ago. Manual review: self-hosted coding sweet spot (32B, 128K, 794k HF dl); superseded by Qwen3-Coder. License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.7/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · HF downloads 793,989 (peer-normalized)
- · released 706 days ago
- · manual review: self-hosted coding sweet spot (32B, 128K, 794k HF dl); superseded by Qwen3-Coder