Qwen3.8 27B batch model listed on OpenRouter

Qwen announced Qwen/Qwen3.8-27B. According to Qwen, the repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format. According to Qwen, these artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, TokenSpeed, etc. According to Qwen, Qwen3.8-27B will be available as a hosted version with more production features, e.g., 1M context length by default, official built-in tools. According to Qwen, the service is coming soon. According to Qwen, Qwen3.8-27B is a native vision-language model that understands images and videos. According to Qwen, Qwen3.8-27B has flexible thinking control with thinking mode on by default and can be disabled per request. According to Qwen, reasoning depth can be tuned with reasoning_effort, and reasoning context from historical messages is retained via preserve_thinking. According to Qwen, model type is Causal Language Model with Vision Encoder. According to Qwen, training stage is Pre-training & Post-training. According to Qwen, number of parameters is 27B. According to Qwen, hidden dimension is 5120. According to Qwen, token embedding is 248,320 (Padded).