Qwen
Qwen is Alibaba Cloud's open-source large language model series, available in sizes from 1.8B to 72B parameters, with both base and chat-aligned variants. Models support 32K context length, multilingual training (Chinese and English focus), and come in multiple quantization formats (Int4, Int8) for reduced memory footprint.
Key facts
Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.
| Field | Value |
|---|---|
| Repository | QwenLM/Qwen |
| Owner | QwenLM |
| Primary language | Python |
| License | Apache-2.0 — OSI-approved |
| Stars | 21.4k |
| Forks | 1.8k |
| Open issues | 41 |
| Latest release | Unknown |
| Last updated | 2026-03-05 |
| Source | https://github.com/QwenLM/Qwen |
What Qwen is
Qwen comprises pretrained base models and instruction-tuned chat models trained on 2.2–3T multilingual tokens, featuring flash-attention optimization, configurable context windows (8K–32K), quantization support (GPTQ, KV cache, Int4/Int8), and tool-use capabilities. Inference memory requirements range from 2.9GB (1.8B Int4) to 48.9GB (72B Int4) for 2048-token generation.
Get the Qwen source
Clone the repository and explore it locally.
git clone https://github.com/QwenLM/Qwen.gitcd Qwen# follow the project's README for install & configurationNeed it deployed, integrated, or customized instead? DEV.co ships production installs.
Best use cases
Implementation considerations
- Select model size (1.8B–72B) and quantization (full-precision, Int8, Int4) based on latency, memory, and accuracy trade-offs; document GPU/CPU requirements upfront.
- Qwen2 is the active successor; evaluate migration path early if planning long-term support—Qwen1 codebase divergence may complicate future transitions.
- Fine-tuning via full-parameter, LoRA, or Q-LoRA; verify Q-LoRA VRAM overhead (5.8GB–61.4GB minimum) aligns with your infrastructure before committing.
- Integrate models via Hugging Face/ModelScope APIs, or build custom OpenAI-compatible API wrappers using provided examples and deployment guides.
- Test context window performance (8K for 14B, 32K for 1.8B/72B) against your workload; long-context evaluation data available in repository.
When to avoid it — and what to weigh
- Require active upstream maintenance and frequent updates — Repository explicitly states 'no longer actively maintained' in favor of Qwen2; expect minimal bug fixes and feature additions to this codebase.
- Need extensive English-language benchmarks and western domain coverage — Model trained with heavy focus on Chinese and English; limited information on performance outside these languages or on specialized western domains.
- Require commercial support or SLA guarantees — Community-supported open-source project; no mention of professional support tiers, incident response, or liability indemnification.
- Cannot tolerate dependency on Alibaba Cloud infrastructure — Primary hosting and API services tied to Alibaba (DashScope API, ModelScope); self-hosting required for air-gapped environments.
License & commercial use
Apache License 2.0 (Apache-2.0). Permissive OSI-approved license allowing commercial use, modification, and distribution with attribution and liability disclaimer. Model weights and code both licensed identically.
Apache-2.0 permits commercial deployment and monetization without royalties. However, verify no export restrictions apply (model is Chinese-origin; check your jurisdiction). Conduct security/bias review before production use. No warranty or indemnification provided; liability rests with deployer. Consider obtaining legal counsel for high-stakes applications.
DEV.co evaluation signals
Editorial assessment — not user reviews. Directional, with an explicit confidence level.
| Signal | Assessment |
|---|---|
| Maintenance | Stale |
| Documentation | Adequate |
| License clarity | Clear |
| Deployment complexity | Moderate |
| DEV.co fit | Good |
| Assessment confidence | High |
No third-party security audit or vulnerability disclosure program mentioned. Model trained on internet-scale data; standard LLM risks (prompt injection, jailbreak) apply. No guardrails mentioned beyond training alignment. Verify output safety for sensitive domains. Self-hosting avoids Alibaba Cloud data exposure, but requires your own infrastructure security hygiene. No rate-limiting or access control documented for inference.
Alternatives to consider
Llama 2 / Llama 3 (Meta)
Larger ecosystem, active maintenance, stronger English performance. Llama 2 Apache-2.0 licensed. Heavier (~70B model) but broader community adoption and tooling.
Mistral 7B / Mixtral (Mistral AI)
Smaller footprint, faster inference, well-maintained. Apache-2.0 licensed. Lower resource overhead and newer architecture, but less mature for Chinese NLP.
Qwen2 (QwenLM/Qwen2)
Official successor to Qwen1; actively maintained, same licensor. Better performance and ongoing support. If starting new project, Qwen2 recommended over this codebase.
Build on Qwen with DEV.co software developers
Evaluate your use case, model size, and infrastructure. Consider Qwen2 for new projects. Start with quantized variants to reduce memory overhead. Test migration if upgrading from Qwen1.
Talk to DEV.coRelated open-source tools
Surfaced by semantic similarity across the DEV.co open-source index.
Related on DEV.co
Explore the category and the services that help you build with it.
Qwen FAQ
Is Qwen production-ready?
Can I use Qwen commercially?
What are the minimum hardware requirements?
Does Qwen support long-context tasks?
Custom software development services
Need help beyond evaluating Qwen? DEV.co is a software development agency offering software development services and web development for teams of every size. Our software developers and web developers build custom software, web applications, APIs, and ai frameworks integrations — and maintain them long-term.
Ready to Deploy Qwen?
Evaluate your use case, model size, and infrastructure. Consider Qwen2 for new projects. Start with quantized variants to reduce memory overhead. Test migration if upgrading from Qwen1.