What Qwen is best at
- The best open-weights quality available today
- Self-hosted production deployments with full control
- Multilingual and multimodal work - exceptional beyond English
Where it isn’t the answer
Consumer chat - the raw models are for builders; everyday users will prefer a polished app.
Plans & pricing
The story so far
- 2023First Qwen releases from Alibaba
- 2024-2025Qwen 2.5 becomes the open-source community favourite
- Aug 3, 2026Qwen3.8-Max previews: 2.4T parameters, 95B active, 1M context
- Aug 12, 2026The Max weights go public - the first open release of a Max-class flagship
- Aug 14, 2026Qwen3.8-27B follows: dense, Apache 2.0, vision and video, 262K context
- Aug 26, 2026Qwen3.8-Flash-Next arrives as the fast, cheap end of the family
Questions people actually ask
Is Qwen open source?
Most of the family ships as open weights. Qwen3.8-27B is Apache 2.0. Qwen3.8-Max is downloadable too, but under a custom Qwen3.8-Max licence rather than a standard open one - read it before shipping commercially.
How good is Qwen really?
Alibaba puts Qwen3.8-Max fifth on Text Arena and second on Vision Arena, which makes it the strongest model anyone has published downloadable weights for. On coding benchmarks specifically, Kimi K3 and GLM from Z.ai trade places with it.
Can I run Qwen at home?
Qwen3.8-27B, yes - it is sized for ordinary on-premise GPUs. Qwen3.8-Max at 2.4 trillion parameters needs a serious cluster.
Who should use Qwen?
Builders. If you are self-hosting, fine-tuning, or need top multilingual performance with full control, it is a first-choice base model.