Overview
Qwen3.8 builds on the Qwen3.5 architecture, and Qwen3.8-27B is its compact, deployment-friendly dense model. It understands images and videos natively and keeps Qwen’s switchable thinking: thinking is on by default, can be disabled per request, and its depth can be tuned with a reasoning-effort setting.
Highlights
- 27B dense parameters
- Native image and video understanding, up to hour-scale videos
- Thinking on by default, with adjustable reasoning effort
- Option to preserve reasoning context across turns
- 262,144-token context window
Typical uses
Local coding agents, document and visual analysis, and multi-step tasks on a single high-memory GPU or workstation.
Things to consider
- Long thinking traces add latency and token cost.
- A hosted version with 1M context and built-in tools is announced on Qwen Cloud as coming soon.
- Download sizes refer to the default quantized versions in the Ollama library; full-precision weights are larger.
Specifications are taken from the model card on Hugging Face, the Ollama library and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.