Qwen3.8-Flash
Language modelOpen weightsCard summarizedThe latest multimodal model from the Qwen family, combining powerful reasoning and generation with remarkable speed
Active (latest Flash)Source: Qwen3.8-Flash-Next model card (Hugging Face) and technical report (28 pp)
Use it via API
Where to call it
- Alibaba Cloud Model Studio
qwen3.8-flash - Self-host (weights)
Qwen/Qwen3.8-Flash-Next
Good for
- Agentic coding at low cost (SWE-bench Pro 62.5%, SWE-bench Multilingual 81.0%)
- Office and professional agent work (CoWorkBench 73.9%, JobBench 55.7%)
- GUI and mobile agents (AndroidWorld 84.5%)
Overview
Qwen3.8-Flash
The latest multimodal model from the Qwen family, combining powerful reasoning and generation with remarkable speed
- Cost
- $0.15 in
- Context
- 1M
- Input / output
- $0.15 / $0.47
- Max output
- 131K
- Knows up to
- —
- Released
- 26 Aug 2026
Qwen3.8-Flash-Next is a 125B-parameter sparse MoE (6B active, plus 51B n-gram embedding parameters held off the accelerator) and the first open release of the architecture planned for Qwen4. It swaps full attention for Qwen Sparse Attention, adds a gated residual stream and n-gram embeddings, and in Qwen's tables beats Qwen3.8-27B and Qwen3.7-Plus on most agent and vision tasks. Qwen3.8-Flash is the API version. model card; tech report abstract
qwen3.8-flash- Input / output per 1M
- $0.15 / $0.47
- Context window
- 1M
- Max output
- 131K
- Knows up to
- —
- Reasoning
- Hybrid (thinking on by default; enable_thinking, thinking_budget; reasoning_effort on the open weights)
- Weights
- Open · Qwen Community License 1.0 (open weights); API proprietary
About this card. A model card is the maker's own account of a model: what it's for, how it was trained and tested, and where it falls short. This page condenses Qwen Team, Alibaba's Qwen3.8-Flash-Next model card (Hugging Face) and technical report. Every claim carries its source page.