The Qwen3.5 native vision-language Flash models are built on a hybrid architecture that integrates a linear attention mechanism with a sparse mixture-of-experts model, achieving higher inference efficiency. Compared to the...
Specifications
- Provider
- Qwen (Alibaba)
- Type
- Vendor / proprietary
- Modality
- text+image+video->text
- Context window
- 1M
- Released
- February 25, 2026
Capabilities
Input: textInput: imageInput: videoOutput: text
Strengths
frequency_penaltyinclude_reasoningmax_tokenspresence_penaltyreasoningresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_p
