Skip to content
← Models
Open weightsAlibaba (Qwen)/10 trustVerified
Qwen2.5-Coder 7B logo

Qwen2.5-Coder 7B

The people's code model — 1.7M downloads of Apache-2.0 7B coding excellence.

Visit official page →

About

Qwen2.5-Coder-7B-Instruct (September 2024) is arguably the most impactful open code model ever released: 1.68M Hugging Face downloads, Apache 2.0, 128K context, and coding quality that embarrassed models 10x its size at release. It became the default local coding model for a generation of developers — the backbone of countless Ollama/vLLM coding setups.

Apache 2.0 licensed — commercial use allowed. Still highly relevant in 2026 for resource-constrained coding assistance, though the Qwen3-Coder family now leads on agentic tasks. If you need one small Apache-2.0 code model that just works, this remains a top answer. The 32B sibling is already in the catalog (qwen2-5-coder-32b).

Specs

Context window

131,072 tokens

Pricing

Free — open weights (Apache 2.0), self-hosted, commercial use allowed.

Free quota

Free — download and run the weights yourself (Apache 2.0); runs on consumer GPUs.

API access

Self-hosted (Transformers, vLLM, Ollama, llama.cpp GGUF); API via Alibaba Cloud and aggregators.

Weights

Open weights

Released

September 19, 2024

Benchmarks

Provider-claimed at release (2024): leading 7B HumanEval/MBPP scores — treat as provider-claimed. Independent: massive community validation; widely benchmarked by third parties with strong results.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

codecodingide-extensionscoding-agentsdebugging

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier). The defining open code model of its era: 1.68M downloads (peer-normalized near the top), Apache 2.0, and third-party-validated quality that held up for two years. Docked two points: (1) the Qwen3-Coder family now leads on agentic coding — this is the efficiency pick, not the capability pick; (2) the 2024 release means the ecosystem has moved to newer architectures.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
9.1/10
Maintenance · recency, releases, issue load
4.0/10
Openness · license + docs
9.4/10
Manual review · sanity read + red-flag check
9.0/10
  • · HF downloads 1,683,947 (peer-normalized)
  • · released 754 days ago
  • · manual review: defining open code model of its era; third-party validated quality; Qwen3-Coder leads on agentic now; license: Apache-2.0

More in the qwen2 family

Sources