Alibaba open-sources Qwen-7B
The 7-billion-parameter model, pretrained on over 2.2 trillion tokens, was released alongside a chat-tuned variant and pitched against Meta's Llama on benchmark scores.
- Open weights & ecosystem
- Models & capabilities
- Minor
Alibaba released open weights for Qwen-7B, a 7-billion-parameter language model pretrained on more than 2.2 trillion tokens, alongside Qwen-7B-Chat, a version fine-tuned on curated instruction and dialogue data. Both were made freely downloadable, and the chat variant supported tool use through ReAct-style prompting, letting it call external functions.
Alibaba’s own reporting claimed Qwen-7B outperformed comparably sized open models — and in some cases larger ones — on benchmarks including MMLU, C-Eval, HumanEval, GSM8K and WMT translation tasks, and coverage framed the release explicitly as a challenge to Meta’s Llama family, which had become the default reference point for open-weight language models since its release that spring.
The release came amid a wave of Chinese labs — Alibaba, Baidu, ByteDance and others — pushing out competing large language models within weeks of each other in mid-2023, part of a broader pattern in which Chinese firms leaned toward open or semi-open releases as a way to build developer adoption quickly, contrasting with the closed-API approach taken by OpenAI and Anthropic. Qwen-7B was an early entry in what became a much larger Qwen model family, later including larger dense and mixture-of-experts variants that Alibaba continued to release openly through the following years.