View on OpenRouter DFlash 2: Keep Drafting Parallel
DFlash 2 is the successor to our widely deployed parallel drafter: close to 3× the speed of autoregressive decoding, with the same output. Drafters for Qwen3.8-27B and Meta's Muse Glimmer are out today.
Inco AI
Qwen3.8 27B
Qwen
Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability.
Specs
Model
- Family
- Qwen
- Variant
- 3.8 27B
- Parameters
- 27B dense
- Weights
- Open
- License
- Apache-2.0
Limits
- Context
- 262,144 tokens
- Max output
- 131,072 tokens
Modalities
- Input
- Text Image Video
- Output
- Text
Capabilities
- Reasoning
- Yes
- Reasoning control
- On / off, Effort levels
- Tool calling
- Yes
Dates
- Released
- 14 Aug 2026
Articles
All articlesinco.ai
Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things
Friday’s big release was Qwen 3.8 27B, an Apache 2 licensed 27B parameter vision-capable LLM from Alibaba’s Qwen research lab. I’ve been looking forward to this one: 27B is an …
Simon Willison
Qwen 27B 3.8: Non-Thinking vs XHigh Thinking
Qwen 27B 3.8 was given the same CRUDbench task in non-thinking mode and XHigh thinking mode.
Lotto@LottoLabs
More from Qwen
All modelsQwen3.8 Flash Next
Qwen
An open-weights multimodal MoE model that doubles as an early preview of the Qwen4 architecture, the same role Qwen3-Next played for Qwen3.5. It pairs Gated DeltaNet with Qwen Sparse Attention, widens the residual stream into four gated branches, and adds 51B N-gram embedding parameters that cost almost nothing per token. Natively 262K context, extensible to 1M with YaRN.
Qwen Community License 1.05w ago