Multiple Models Per Version: Train and compare models side by side

You can now train multiple models on the same dataset version. Kick off a training on a version, then start a second one in parallel: train every size of YOLOv8, YOLO26, and RF-DETR on identical data and compare the results directly.

Models trained after the rollout register with a new full model ID, shown on the model page, while existing project and version references keep working unchanged. The MCP Server supports the new capability too, so an agent can launch and compare trainings for you.