Nvidia open-sources Flora, cutting robot video labeling cost 125x

Nvidia's Flora data engine labels an hour of robot video for 72 cents, versus $88 with GPT-6 Astra, and scores slightly higher on quality. Built on three fine-tuned Qwen3.5 models, it catches errors in robot actions more reliably than competitors but sometimes trails on descriptive accuracy.

Why it matters: Labeling is a major bottleneck for robot training data, and a two-orders-of-magnitude cost drop makes large video datasets practical for smaller teams. It also fits Nvidia's push to own more of the robotics stack, from chips to data tooling.