Nvidia open-sources Flora for robot video labeling
Original title: Nvidia open-sources Flora to make robot video labeling 125x cheaper than GPT-6 Astra
Nvidia says three fine-tuned Qwen3.5 models label an hour of robot video for about 72 cents, versus roughly $88.50 with GPT-6 Astra.
Nvidia's GEAR lab released Flora, an open data engine that segments robot videos, captions actions, and scores success, step quality, and outcomes. Nvidia says it already annotates up to 200,000 hours a day for GR00T training on 512 GB300 GPUs, and the models are on Hugging Face. On Nvidia's tests Flora beats GPT-6 Astra on some localization and error-detection tasks but trails on others; the cost figures are projections, not production measurements.
Unconfirmed: Cost and quality figures are Nvidia's own estimates and benchmarks; GPT-6 Astra also judged some descriptions while being a model under test.
- Flora uses a 9B segmenter and two 27B Qwen3.5 models, including Flora-Rewarder.
- Nvidia estimates 57.1 percent quality at $0.72 per video hour versus 53.7 percent at $88.49 for GPT-6 Astra.
- Costs assume a $10 per GPU-hour rental price.
- Flora-Rewarder catches 53.3 percent of failed steps; no competitor exceeds 11.7 percent, Nvidia says.
- Flora falsely flags about 11 percent of normal steps as errors.
From the original article
4 min read at the-decoder.com
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.
By Jonathan Kemper · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.