Voxel51 3.0 unifies search, curation, annotation and evaluation into an end-to-end multimodal data platform.

SAN FRANCISCO, Oct. 15, 2026 — Voxel51, the data platform behind AI systems at some of the world’s leading companies, today announced Voxel51 3.0., a major expansion of its platform for building and managing physical AI datasets.

The next wave of AI is moving from the internet into the physical world. In particular, robotics is entering a phase that resembles the early days of autonomous vehicles: rapid investment, increasingly sophisticated foundation models, and an urgent need for infrastructure built around complex real-world data.

Voxel51 CEO Brian Moore adds, “We saw this play out with our customers building autonomous systems and perception models. The teams that got their data infrastructure right from the start were able to move faster. Physical AI and robotics is at that same inflection point. The companies building that foundation today are going to have a real advantage as their data and models scale.”

Voxel51 3.0 brings the core data workflows for physical AI into a single, extensible platform built for teams moving from prototype to production. New capabilities include:

• Voxel51 Agent: An AI agent that works across the physical AI data loop to search, curate, label, evaluate, and chain actions into repeatable workflows.
• Voxel51 Search: A persistent, hybrid search index that combines semantic search, lexical queries, and metadata filters as datasets grow.
• Voxel51 Annotation: Annotation workflows across 2D, 3D, video, and multimodal data, including automated labeling with vision-language models through Agentic Labeling.
• Multimodal time-series support: Synchronized visualization of sensor streams and segment-level embeddings for finding specific moments within long recordings.
• Data provenance: Activity logs, sample edit history, and dataset versioning that track changes and help teams identify exactly what data went into a model.
• Bring your own models: A model-agnostic architecture that supports foundation models and domain-specific models across embeddings, inference, auto-labeling, and evaluation.
• Free self-service access: The ability to get started with Voxel51 for free at app.voxel51.com

Voxel51 is built for the physical AI data flywheel

Models are getting easier to build. Data work is not. And teams that consistently get models into production invest heavily in the work of systematically finding, fixing, and improving their training data. According to the State of Physical AI 2026 report, exceptional teams are three times more likely than struggling teams to spend more than half their project time on data work.

Voxel51 3.0 brings the core workflows for physical AI into an end-to-end data platform, helping teams move from raw multimodal recordings to searchable, curated training data and support active learning as models improve.

Voxel51 Agent is the most capable agent for the physical AI data flywheel. The agent draws on more than 45 MCP tools, 80+ built-in operations, and dozens of skills to chain actions across search, curation, labeling, evaluation, and other data workflows.

“Physical AI teams don’t need another chat interface sitting on top of their data,” said Voxel51 Product Manager Adonai Vera. “They need an agent that can maintain context across the model development lifecycle and orchestrate actions across the full data flywheel.”

Nebius and Voxel51 partner to connect the physical AI data and compute stack

Nebius and Voxel51 address two of the hardest infrastructure problems in physical AI: compute and data.

Nebius provides predictable performance and reliable cloud infrastructure as an integrated system engineered for AI, while Voxel51 helps teams turn large volumes of raw recordings into model-ready datasets.

Together, the companies are hosting a series of hands-on workshops where developers can run Voxel51 workflows on Nebius AI Cloud. The goal is to help teams build a tighter physical AI data flywheel while spending less time stitching together the underlying data and compute stack.

“Physical AI is the next major frontier for AI because it brings digital intelligence into the world of atoms,” said Evan Helda, Head of Physical AI at Nebius. “But getting there will take more than better models. Robotics teams need the compute, cloud platform, and data infrastructure to turn every deployment, failure and edge case into a faster next iteration. That’s the kind of foundation we’re building with partners like Voxel51.”

About Voxel51

Voxel51 is an end-to-end data platform for physical AI, helping teams turn complex multimodal datasets into better AI models. The platform unifies search, visualization, curation, annotation, and evaluation, with AI-powered workflows that automate data-intensive tasks, while giving developers control over their models and infrastructure. Voxel51 is trusted by teams at Berkshire Grey, Ford, Porsche, and Precision Planting to develop AI systems that operate in the physical world.

SOURCE Voxel51