Typed answers your code can act on — probabilities, choices, and scores, not prose.
★★★★☆4.3
N
NVIDIA Flora
Open robotics data engine that labels robot video at a fraction of frontier-model cost.
Try it Free →
♡ 3,234 saves
What is NVIDIA Flora?
NVIDIA Flora is the robotics data engine behind NVIDIA's GR00T humanoid models, open-sourced by NVIDIA's GEAR lab in October 2026. Robot training needs video labeled with what happened, when, and how well it went — and Flora covers all three in one suite: Flora-TempSeg (9B) splits a recording into action intervals, Flora-Captioner (27B) writes one short instruction per interval, and Flora-Rewarder (27B) scores each episode's outcome, classifies steps as normal, suboptimal, or failed, and grades against rubrics you write.
NVIDIA says Flora already labels 200,000 hours of video a day inside GR00T training and puts the modeled cost at about $0.72 per hour of video versus roughly $88.49 for GPT-6 Astra — a company estimate based on assumed GPU rental pricing, not a measured bill. The weights are free to download on Hugging Face under NVIDIA's OneWay Noncommercial license, which makes Flora a genuinely useful research tool for robotics teams while ruling out commercial deployment.
Key Features
- Three models in one suite: TempSeg 9B segments actions, Captioner 27B describes them, Rewarder 27B scores them
- Timestamped action intervals exported as JSON for training pipelines
- Outcome scores, dense step-by-step rewards, and custom rubric scoring in a single model
- The same data engine NVIDIA says powers GR00T-2 training at massive scale
- Weights available on Hugging Face under NVIDIA's noncommercial license
Pros
- Purpose-built for robot training data instead of repurposing a general chatbot
- Covers segmentation, captioning, and reward scoring in one suite rather than three tools
- Modeled labeling cost is a small fraction of frontier API pricing for the same job
Cons
- Noncommercial license — research and evaluation only, with commercial and production use explicitly unsupported
- The headline cost and quality figures are NVIDIA's own estimates, and benchmarks are mixed across tasks
- Running the 27B models takes serious GPU hardware; the 200,000-hour daily throughput figure reflects NVIDIA's own 512-GPU deployment
Pricing
- Model: Free
- Free to download on Hugging Face for non-commercial research under NVIDIA's OneWay Noncommercial license; commercial and production use explicitly unsupported — check the model cards for current terms.
Adding Up Your AI Bill?
If you pay for NVIDIA Flora alongside other AI subscriptions, the monthly total adds up quietly. Tick every tool you pay for in our free AI Tool Cost Calculator to see your real monthly and yearly bill — and spot overlaps you can cut. Still choosing? Start with our guide on how to choose the right AI tool.
How We Rated It
Scores reflect our editors’ independent assessment at the time of review.
Alternatives to NVIDIA Flora
Open-source desktop AI agent — bring your own LLM, skip the vendor lock-in.
★★★★☆4.0
Turn any website into an API your AI agent can call — no API required.
★★★★☆4.2