Open-weight dense 27B vision-language model from Qwen for long-running agent tasks.
Add a second model to compare.