It's ok, somewhere between a qwen 2.5 VL and the frontier models (o3 / opus 4) on visual reasoning