Qwen (Alibaba) released Qwen3.8-27B on Hugging Face under Apache-2.0 license. The model is a 27B-parameter native vision-language model with 262,144 native context length, extensible to 1,000,000 tokens. It supports flexible thinking control and is compatible with Hugging Face Transformers, vLLM, SGLang, and TokenSpeed. Qwen Cloud API service is coming soon.
China context
- Original name
- 阿里巴巴
- Outside China
- Open weights · huggingface.co
- Claims
- Company-reported; not yet independently evaluated
- For builders
- Developers outside China can download and deploy Qwen3.8-27B under Apache-2.0, with compatibility across major inference frameworks.
- For investors
- The release indicates Alibaba's commitment to open-weight models and a forthcoming managed API, which may affect competitive dynamics in the AI model market.
Qwen released Qwen3.8-27B, a 27B-parameter open-weights vision-language model, on Hugging Face under Apache-2.0. It features native image and video understanding, flexible thinking control, and a 262,144-token native context length extensible to 1M. The model is compatible with major inference frameworks and a hosted Qwen Cloud API is announced as coming soon.
Qwen3.8-27B uses a hybrid architecture with 64 layers, alternating Gated DeltaNet and Gated Attention blocks, and multi-token prediction training. It has 248,320 padded token embeddings and supports reasoning effort tuning and preserve thinking for historical context retention.
The release continues Qwen's pattern of open-weight model releases with permissive licensing, targeting developers and enterprises seeking deployable vision-language models. The announced Qwen Cloud API indicates a dual open-source and managed service strategy.
For builders, Qwen3.8-27B offers a deployable open-weights vision-language model with flexible thinking control and long context. For investors, it signals Alibaba's continued investment in open-model releases and API monetization.
Observable next signals include the launch of the Qwen Cloud hosted version with 1M context and built-in tools, community adoption metrics on Hugging Face, and independent benchmark evaluations.