Qwen 3.8 27B model release
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
Alibaba released Qwen3.8-27B, a 27-billion-parameter open-weights multimodal AI model. It outperforms larger rivals on coding and agentic tasks, signaling a shift toward efficient local AI.
Background
Alibaba's Qwen team released the open weights for Qwen3.8-27B, a dense multimodal model with 27 billion parameters, under the Apache 2.0 license. The model is positioned as a compact, deployment-friendly option that competes with larger models like Qwen3.7-Plus and Muse Glimmer-30B on coding and office tasks. It supports native image and video understanding, a flexible thinking mode, and a context window of 262K tokens extendable to 1M via YaRN. The release follows a promise of open weights and includes a larger Max-class model, Qwen3.8-2.4T-A95B.
How it unfolded
- Aug 13, 2026AMD announced Day-0 support for Qwen 3.8 on Instinct GPUs, enabling immediate deployment with SGLang and vLLM.
- Aug 14, 2026Alibaba's Qwen team released open weights for Qwen3.8-27B and Qwen3.8-2.4T-A95B on Hugging Face and ModelScope.
- Aug 15, 2026AMD announced Day-0 support for Qwen3.8-27B on Ryzen AI Max PCs and Radeon GPUs, and media coverage highlighted benchmark results.
Who’s saying what
- Official
- Qwen team states Qwen3.8-27B outperforms Qwen3.7-Plus in coding and office tasks, and is built for builders with efficient local deployment.
- Analysts
- Tech analysts note Qwen3.8-27B achieves state-of-the-art performance for its size, comparable to Opus4.6 Max on several benchmarks.
Still unverified
Benchmark comparisons are based on Qwen's own reported numbers and have not been independently verified.