Qwen3.6-27B FP8 Released

RedHatAI released Qwen3.6-27B-FP8 on Hugging Face.

• It is the first open-weight variant of Qwen3.6, emphasizing agentic coding and thinking preservation.
• It provides FP8-quantized weights and configuration files in the Hugging Face Transformers format.
• It is compatible with Transformers, vLLM, SGLang, and KTransformers, and applies fine-grained FP8 quantization with a block size 128.
• Performance metrics are described as nearly identical to the original model, and the model card presents 27B parameters, a default context of 262,144 tokens, and an extension of up to 1,010,000 tokens.