The default nvidia-vision model alias should point to nvidia/nemotron-3-nano-omni-30b-a3b-reasoning instead of the older nvidia/nemotron-nano-12b-v2-vl. The new model has better multimodal capabilities and should use inference parameters temperature=0.60, top_p=0.95.

Users with an existing ~/.data-designer/model_configs.yaml will need to delete it for the new default to take effect.