Qwen is an 📝Alibaba product — a family of predominantly open-weight 🏷️#large-language-models and multimodal models built by the Qwen team at Alibaba Cloud, also known by its Chinese name Tongyi Qianwen (通义千问). Alibaba released the first beta in April 2023 and opened public access that September after regulatory approval; successive generations — Qwen2 (June 2024), Qwen2.5 (September 2024), the QwQ reasoning model (late 2024), Qwen3 (April 2025), Qwen3.5 (February 2026), Qwen3.6 (April 2026), and 📝Qwen3.8 (August 2026) — made it one of the most widely adopted open-weight lineages, with more than 200,000 derivative models on Hugging Face. It serves developers who self-host or fine-tune checkpoints, enterprises calling hosted models through Alibaba Cloud Model Studio, and consumers through the Qwen app, which Alibaba reported at 234 million users by May 2026 and which connects to Taobao, Fliggy, and food delivery.
Key Features
- Open-weight releases — Most generations ship downloadable checkpoints under Apache 2.0 or Qwen licenses, spanning sub-billion-parameter models up to trillion-parameter mixture-of-experts systems.
- Hybrid thinking modes — Since Qwen3, models switch between step-by-step reasoning and fast direct answers, with reasoning effort controllable instead of fixed per model variant.
- Multimodal lines — Qwen-VL, Qwen-Audio, and Omni variants extend the family to vision, audio, and video input; 📝Qwen3.8-Max folds visual understanding into the flagship.
- Agentic coding — Qwen-Coder models and the Qwen Code CLI target autonomous software engineering; Alibaba reports Qwen3.8-Max ran a 16-day unattended build of an agent framework.
- QwenWork — Alibaba's all-in-one workplace AI agent platform, announced in August 2026 as a way to use Qwen3.8-Max outside the API and the Model Studio developer console.
Getting Started
- Try the current models free in Qwen Chat on the web or in the Qwen mobile app.
- Call hosted models through the Alibaba Cloud Model Studio API, which is 📝OpenAI-compatible.
- Download open weights from the Qwen organization on Hugging Face or on ModelScope.
- Run smaller checkpoints locally through 📝Ollama, vLLM, or SGLang on your own hardware.
Related
- 📝Open-Weight vs. Closed-Weight AI Models — the strategic tradeoffs behind Qwen's distribution model