Qwen3.8 27B
Qwen's dense 27B vision-language model for coding, research, and long-horizon agents with controllable thinking and native speculative decoding through MTP
Serve the model
Pick your Jetson, then copy the command and paste it into a terminal on the Jetson. New to Jetson? Setup guide
Model Details
View on HuggingFace
Qwen3.8 27B is Qwen’s dense, open-weight vision-language model for coding, professional work, research, and long-horizon agentic tasks. It brings the strongest generation of Qwen open models to a deployment-friendly size, with better planning and stronger handling of tool and environment feedback for more reliable multi-step task completion.
Thinking is enabled by default and can be disabled per request. Reasoning depth is adjustable with xhigh, medium, and low effort levels, while preserved thinking carries reasoning context across turns. The model also supports a native 262K context window and is trained with multi-step MTP, which the Jetson commands enable for faster generation.
Modalities
Input: Text, image, and video
Output: Text
Performance on Jetson
- 12.5 vLLM on AGX Thor: 12.5 tok/s 8.2 llama.cpp on AGX Thor: 8.2 tok/sAGX Thor128 GB
- 7.7 vLLM on AGX Orin: 7.7 tok/s 5.3 llama.cpp on AGX Orin: 5.3 tok/sAGX Orin64 GB
Output tokens/s
- 54.9 vLLM on AGX Thor: 54.9 tok/s 14.3 llama.cpp on AGX Thor: 14.3 tok/sAGX Thor128 GB
- 20.5 vLLM on AGX Orin: 20.5 tok/s 9.9 llama.cpp on AGX Orin: 9.9 tok/sAGX Orin64 GB
Output tokens/s
- 25.9 MTP vLLM with MTP on AGX Thor: 25.9 tok/s 31.4 DFlash2 vLLM with DFlash2 on AGX Thor: 31.4 tok/s 30.6 DSpark vLLM with DSpark on AGX Thor: 30.6 tok/s 17.2 MTP llama.cpp with MTP on AGX Thor: 17.2 tok/sAGX Thor128 GB
- 14 MTP vLLM with MTP on AGX Orin: 14 tok/s 17.6 DFlash2 vLLM with DFlash2 on AGX Orin: 17.6 tok/s 17.9 DSpark vLLM with DSpark on AGX Orin: 17.9 tok/s 11.6 MTP llama.cpp with MTP on AGX Orin: 11.6 tok/sAGX Orin64 GB
Output tokens/s
- 129 MTP vLLM with MTP on AGX Thor: 129 tok/s 134 DFlash2 vLLM with DFlash2 on AGX Thor: 134 tok/s 130 DSpark vLLM with DSpark on AGX Thor: 130 tok/s 35.3 MTP llama.cpp with MTP on AGX Thor: 35.3 tok/sAGX Thor128 GB
- 56 MTP vLLM with MTP on AGX Orin: 56 tok/s 52.3 DFlash2 vLLM with DFlash2 on AGX Orin: 52.3 tok/s 47.3 DSpark vLLM with DSpark on AGX Orin: 47.3 tok/s 24.2 MTP llama.cpp with MTP on AGX Orin: 24.2 tok/sAGX Orin64 GB
Output tokens/s