Nvidia’s Flora Makes Labeling Robot Videos Cheaper—With One License Caveat

🔍 Read the full analysis: Nvidia’s Flora Makes Labeling Robot Videos Cheaper—With One License Caveat on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Nvidia’s GEAR robotics lab has released Flora, a three-model suite that segments, captions and scores robot videos, and estimates it can label video for $0.72 per hour. The figure is a modeled GPU cost, not a measured production bill, and the models’ noncommercial license permits research or evaluation only, limiting their use as a free labeling service for businesses.

Flora’s models target different parts of the labeling process. Flora-TempSeg, with 9 billion parameters, identifies the time spans associated with separate actions. The 27-billion-parameter Flora-Captioner writes an English description for each segment, while the 27-billion-parameter Flora-Rewarder scores an episode, grades short steps as normal, suboptimal or failure, or evaluates performance against a user’s rubric. Nvidia says the models were built on Alibaba’s Qwen3.5 models.

The lab’s cost comparison uses 50 episodes totaling 4,483 seconds—about 75 minutes—from Nvidia’s DenseCaptionBench. The paper reports GPT-6 Astra API usage at $88.49 per hour of video, using public rates cited as of October 4, 2026. Flora’s $0.72 estimate assumes eight GB300 GPUs rented at $10 per GPU-hour and modeled throughput. Nvidia gives a range of $0.51 to $1.26 per video hour, excluding storage and overhead; the paper says the production GPU-hour denominator has not been verified. The comparison covers captioning, not the reward model.

On the same set, the paper reports a quality score of 57.1% for Flora and 53.7% for Astra. Those results do not mean Flora wins every task: the paper’s broader benchmarks show stronger Flora results on some human-video and partial-success cases, while Astra performs better on robot videos, caption accuracy and rubric scoring. Nvidia ran most of the tests itself, and the comparison is based on a limited set of episodes.

Lower Labeling Costs, With Limits

Video labels help robot developers identify what happened, when an action began or failed, and how well a task was completed. Those annotations can support training-data selection and reinforcement learning. If a lower-cost labeling method performs adequately at scale, it could reduce one expense in building robot datasets, particularly when teams need to process large volumes of footage.

But the headline cost gap needs careful reading. $0.72 is a model-based estimate using an assumed GPU rental rate and throughput; Astra’s figure is reported API usage. The test involved roughly 75 minutes of video selected for Nvidia’s benchmark, not a long-running production deployment. The estimate also excludes storage and overhead and does not include all three Flora models. It points to a potential cost advantage, not proof of a particular company’s savings.

The license is a separate practical constraint. Nvidia’s OneWay Noncommercial License allows research or evaluation, and derivatives carry the same restriction, according to the source material. Companies may examine Flora or use it for permitted evaluation, but cannot treat the released models as a commercial production pipeline under that license. Nvidia’s own commercial use does not change the terms applying to other users.

Three Models for Robot Video

Robot-learning systems are trained using recordings of tasks such as placing an object or folding fabric. The Flora paper describes three labeling needs: timestamped captions that describe actions, step-level process rewards that mark how an action went, and outcome rewards that score an entire attempt. These signals let developers sort training data and examine where a robot’s attempt succeeded or failed.

Nvidia says Flora served as a data engine for its GR00T-2 robot models. The model cards say Flora was trained on fewer than 10,000 hours of human and robot video. The models and inference code are available through Hugging Face repositories, according to the supplied material. The two 27-billion-parameter models are resource-intensive: the source estimates about 55 GB for each model’s weights at 16-bit precision, pointing to data-center hardware rather than a typical personal computer.

The cost claim is calculated by dividing Astra’s reported $88.49-per-video-hour API cost by Flora’s modeled $0.72 estimate, yielding 122.5, rounded by Nvidia to approximately 125 times. Nvidia’s paper also gives Flora cost scenarios of $0.51 to $1.26 per hour. Those calculations use the same benchmark episodes, but compare an API bill with a modeled GPU budget, so the ratio is not a direct comparison of two measured production costs.

““approximately 125×””

— Nvidia, in its description of the cost comparison

Cost and Performance Still Need Testing

The supplied material does not establish what Flora costs under a real production workload. The estimate depends on assumed GPU pricing and throughput, and Nvidia says the production GPU-hour denominator has not been verified. Actual costs could vary with hardware access, video characteristics, workload scale, storage, engineering and other overhead.

How well Flora generalizes beyond the benchmark also remains uncertain. The reported comparison uses 50 episodes from Nvidia’s own test set, and most evaluations were run by Nvidia. The results are mixed across tasks: Flora scores higher on the reported overall test figure but does not lead on every robot-video, captioning or rubric-scoring measure. Independent evaluations and longer production trials are not provided in the source material.

The license terms leave commercial deployment constrained, but the material does not detail every possible use case or how the terms would apply to a particular company’s workflow. Businesses should consult the license itself before relying on the models; the announcement does not describe a separate commercial license or pricing option.

Independent Tests and License Terms

The immediate next step for researchers is to test Flora on additional datasets and hardware configurations, while tracking both labeling quality and the full operating cost. Comparisons that measure GPU usage directly over sustained workloads would help establish whether the modeled savings hold beyond Nvidia’s selected episodes. Tests should also check whether the models’ different strengths matter for specific tasks, rather than treating one benchmark score as a universal measure.

Organizations considering Flora will need to review the OneWay Noncommercial License before using the models, including for any planned evaluation, modification or deployment. The source material identifies research and evaluation as permitted purposes but does not announce commercial terms. No independent production-cost result or license change is reported here; those remain open developments.

Key Questions

What is Nvidia Flora?

Flora is a suite of three models for robot-video labeling: one segments actions by time, one captions those segments, and one scores steps or full episodes.

Is the $0.72-per-hour figure a measured production cost?

No. It is a modeled estimate based on eight GB300 GPUs, an assumed rental price of $10 per GPU-hour and modeled throughput. Nvidia’s paper says the production GPU-hour denominator has not been verified.

Does Flora outperform GPT-6 Astra?

The reported results are mixed. Flora scored 57.1% versus Astra’s 53.7% on the 50-episode comparison, but the paper reports Astra leading on some robot-video, caption-accuracy and rubric-scoring tasks.

Can companies use Flora commercially?

The supplied material says the models use Nvidia’s OneWay Noncommercial License, which allows “research or evaluation purposes only,” with derivatives subject to the same restriction. It does not report separate commercial terms.

What does the approximately 125-times cost comparison include?

It compares Astra’s reported API cost of $88.49 per video hour with Flora’s modeled $0.72-per-hour GPU estimate on 50 benchmark episodes. The comparison covers captioning, excludes storage and overhead, and does not include Flora’s reward model.

Source: ThorstenMeyerAI.com