Nvidia open-sources Flora to make robot video labeling 125x cheaper than GPT-6 Astra
Nvidia has released Flora, an open-source data engine for labeling robot video, according to The Decoder. The company says labeling one hour of robot footage costs 72 cents with Flora, compared with $88 using GPT-6 Astra, a roughly 125x reduction, and that Flora also edges out that competitor on quality scores.
Flora is built on three fine-tuned Qwen3.5 models rather than a single general-purpose model. According to the report, it identifies errors in robot actions far more reliably than any competitor, though it sometimes falls short on descriptive accuracy. That trade-off suggests Flora is tuned for spotting where a robot went wrong rather than for richly narrating what it did.
Nvidia is positioning Flora as a core part of its push to control the entire robotics value chain, per The Decoder. This suggests the release is as much a strategic play as a tooling one: cheaper, higher-quality labeled video lowers the cost of building robot-learning datasets, which in turn feeds demand for the compute and software Nvidia already sells.
Why it matters: Teams training robot policies often spend heavily on video annotation, so a 125x cost drop on a claimed higher-quality pipeline could change dataset economics, though the quality results are Nvidia's own. My inference is that open-sourcing Flora also pulls robotics developers deeper into Nvidia's stack.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Nvidia's new data engine Flora labels one hour of robot video for 72 cents instead of $88 with GPT-6 Astra, while scoring slightly higher on quality. The open system, built on three fine-tuned Qwen3.5 models, catches errors in robot actions far more reliably than any competitor but sometimes falls short on descriptive accuracy. Nvidia is positioning Flora as a key piece of its strategy to control the entire robotics value chain. The article Nvidia open-sources Flora to make robot video labeling 125x cheaper than GP