Voxel51 3.0 unifies the multimodal data loop for physical AI in one platform

What is changing?

Voxel51 announced Voxel51 3.0 for physical AI teams on October 15. The release combines multimodal data search, curation, annotation, and model evaluation in one extensible platform. Time-series support is designed to visualize multiple sensor streams on the same timeline and find specific moments inside long recordings through segment-level embeddings.

Unified annotation workflows cover 2D, 3D, video, and multimodal data. Agentic Labeling uses vision-language models to generate labels from natural-language instructions, while Voxel51 Agent chains search, curation, annotation, and evaluation into repeatable workflows. The company says the agent layer includes more than 45 MCP tools, over 80 operators, and 18 open-source skills.

The release also adds data provenance through activity logs, sample edit history, and dataset versioning. Teams can bring their own models for embeddings, auto-labeling, inference, and evaluation. Voxel51 3.0 is available to try for free with a team’s own data.