Alibaba

AI/ML · Open-Source Software

Launch

Alibaba launches laptop-ready Qwen3.8-27B, rivaling Meta

Open-source developer adoption, not raw benchmark scores, is emerging as the real currency in the global AI model race.

• Alibaba released Qwen3.8-27B, a 27-billion-parameter open-weight model built to run locally on high-end laptops, and open-sourced the full weights of its flagship Qwen3.8-Max model.
• Alibaba's Qwen family already leads open-weight AI adoption: models built on Qwen generated roughly 151,448 derivative projects on Hugging Face, about 2.6 times Meta's tally.
• Qwen3.8-27B natively handles 262,000 tokens (up to 1 million with YaRN) and, per Alibaba's benchmarks, matches models ten times its size, beating Meta's rival laptop model Muse Glimmer-30B on tests like TerminalBench2.1 (73.0 vs 51.7).
• Qwen3.8-Max scales to 2.4 trillion total parameters with 95 billion active parameters, the first time Alibaba has open-sourced a model at that scale, though its training data and methods stay undisclosed.
• The release comes days after Meta unveiled its own laptop-ready open model, Muse Glimmer, part of a push to reclaim ground from Chinese open-weight labs including Alibaba, DeepSeek, and Moonshot.
• On-device AI is becoming the next major battleground since local models offer faster responses and better data privacy, and the speed of Alibaba's counter to Meta shows how quickly the open-weight rivalry is escalating.