Qwen opens a 27B vision-language agent model

Alibaba’s Qwen has released Qwen3.8-27B, an open-weight vision-language model with native image and video understanding, 262,144-token context that can extend to 1 million tokens, and adjustable reasoning effort. The Hugging Face model card reports 61.7% on SWE-bench Pro, 70.7% on CoWorkBench, and compatibility with vLLM, SGLang, and TokenSpeed; it also says thinking can be disabled or tuned per request. Teams evaluating open models should compare this compact deployment option with the earlier Qwen3.8-Max preview’s nine-H200 requirement, because model size—not just benchmark rank—is becoming an infrastructure decision.