Qwen 3.8 27B is out: open weights, best local dense model yet
Summary
Qwen3.8-27B-FP8 is released as a FP8-quantized 27B dense vision-language model on HuggingFace, with an Apache 2.0 license and deployment guidance for Transformers, vLLM, SGLang, and Docker. The model boasts a native context length of 262,144 tokens, expandable to 1,000,000 via YaRN, and targets robust performance for coding, office tasks, and long-horizon workflows. The release highlights open weights and a host of quickstart resources, including model cards and API usage instructions.