March 8, 2026

Qwen (通义千问, Tongyi Qianwen — roughly “seeking answers from all things”) is Alibaba Cloud’s large language model family. It has grown remarkably fast from a quiet internal project to one of the most capable open-weight model series in the world.

The Beginning (2023)

Alibaba Cloud first unveiled Tongyi Qianwen in April 2023 at the Alibaba Cloud Summit, initially as a closed beta integrated into enterprise tools. The model was positioned to compete with ChatGPT in the Chinese market, but Alibaba’s ambitions extended further.

In September 2023, Qwen 1.0 was released, followed shortly by open-weight versions on Hugging Face. The release included 7B and 14B parameter models and was notable for strong performance in both Chinese and English, along with support for a long 8K context window, unusual for models of that generation.

Rapid Iteration (2024)

2024 was the year Qwen went from regional contender to global player.

• February 2024 — Qwen 1.5: A major quality jump across all sizes (0.5B to 72B), with improved instruction following and multilingual support. The 72B model benchmarked competitively with models twice its size.
• June 2024 — Qwen 2: Expanded the lineup to include a 57B mixture-of-experts variant and a 72B dense model. Qwen2-72B topped several open-source leaderboards at release.
• September 2024 — Qwen 2.5: Released alongside a full suite of specialized models — Qwen2.5-Coder for programming tasks and Qwen2.5-Math for mathematical reasoning. The flagship 72B model matched or exceeded GPT-4o on several benchmarks.
• November 2024 — QwQ-32B-Preview: Alibaba’s answer to the reasoning model trend sparked by OpenAI’s o1. QwQ (a playful, questioning name) applied extended chain-of-thought reasoning, delivering impressive results on math and logic tasks despite being a preview release.

Into 2025

• January 2025 — Qwen 2.5 Max: A closed, API-only model that Alibaba claimed outperformed DeepSeek V3 and GPT-4o on key benchmarks. It signaled Alibaba’s intent to compete at the frontier, not just in open weights.
• March 2025 — QwQ-32B (full release): The reasoning model matured into a full release, competitive with much larger models and standing out for efficiency.
• April 2025 — Qwen 3: A new generation bringing native hybrid reasoning (combining fast and slow thinking modes), extended context up to 128K tokens, and a suite ranging from 0.6B to 235B parameters (including MoE variants). Qwen3-235B-A22B became one of the top-ranked open-weight models globally.

What makes Qwen interesting is the open-weight strategy. Alibaba releases most of its models permissively, enabling fine-tuning and local deployment. This has made Qwen models a popular base for researchers and companies worldwide.