NVIDIA releases Flora data engine
Overview
NVIDIA has released 'Flora', a suite of open AI models designed to automatically generate robot learning data. Flora includes three models for action segmentation, captioning, and reward generation, and is claimed to outperform other models in certain benchmarks while being significantly more cost-effective for large-scale video processing.
From Robotics — Robot learning news discovery
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · data annotation hours per day: 200000 hours · Basis not reported
“目前，该数据引擎依托 512块GB300 GPU， 每天可标注 20万小时的数据。”
Exact source · revision 2Source owner not reported
“目前，该数据引擎依托 512块GB300 GPU， 每天可标注 20万小时的数据。”
Exact source · revision 2Source owner not reported
Reported quantity · cost per hour of source video: 0.72 other · Basis not reported · Differing source assertions
“具体来看，Flora的标注质量得分为 57.1% ，估算成本为 每小时源视频0.72美元； GPT-6 Astra的标注质量得分为 53.7%， 估算成本为每小时源视频 88.49美元。”
Exact source · revision 2Source owner not reported
“具体来看，Flora的标注质量得分为 57.1% ，估算成本为 每小时源视频0.72美元； GPT-6 Astra的标注质量得分为 53.7%， 估算成本为每小时源视频 88.49美元。”
Exact source · revision 2Source owner not reported
Reported quantity · cost per hour of source video: 88.49 other · Basis not reported · Differing source assertions
“具体来看，Flora的标注质量得分为 57.1% ，估算成本为 每小时源视频0.72美元； GPT-6 Astra的标注质量得分为 53.7%， 估算成本为每小时源视频 88.49美元。”
Exact source · revision 2Source owner not reported
“具体来看，Flora的标注质量得分为 57.1% ，估算成本为 每小时源视频0.72美元； GPT-6 Astra的标注质量得分为 53.7%， 估算成本为每小时源视频 88.49美元。”
Exact source · revision 2Source owner not reported
Reported quantity · Reported accuracy: 64.7 percent · Basis not reported
“同时，使用公开的纯文本动作描述评分标准，联合匹配则要求预测的动作时间区间，和动作描述均与标注结果相符 Flora的平均分段F1为 50.3%， 平均联合F1为 34.5%， 在人工标注参考区间上的动作描述准确率为 64.7%。”
Exact source · revision 2Source owner not reported
“同时，使用公开的纯文本动作描述评分标准，联合匹配则要求预测的动作时间区间，和动作描述均与标注结果相符 Flora的平均分段F1为 50.3%， 平均联合F1为 34.5%， 在人工标注参考区间上的动作描述准确率为 64.7%。”
Exact source · revision 2Source owner not reported
Reported quantity · video hours processed per day: 200000 hours · Basis not reported
“at approximately 1/125th the cost of GPT-6 Astra. Furthermore, it has a track record of labeling 200,000 hours of video per day using 512 GB300 GPUs. This demonstrates that robot learning data can be generated not just at the laboratory level, but at the scale of large data centers. However, these costs and processing speeds are based on NVIDIA's measurement conditions. 6. Model release The pre-trained weights for Fl”
Exact source · revision 2Source owner not reported
Reported quantity · cost per hour of video: 0.72 other · Basis not reported · Differing source assertions
“criteria. These are the benchmark results reported by NVIDIA. 5. Processing cost and scalability A major feature of Flora is that it achieves evaluation performance comparable to frontier models at a low cost. The labeling cost per hour of video estimated in the paper is as follows: ・Flora: $0.72 ・GPT-6 Astra: $88.49 ・Claude Opus 5: $41.70 ・Gemini 3.8 Flash: $8.48 Under NVIDIA's evaluation conditions, Flora operates”
Exact source · revision 2Source owner not reported
Reported quantity · cost per hour of video: 88.49 other · Basis not reported · Differing source assertions
“criteria. These are the benchmark results reported by NVIDIA. 5. Processing cost and scalability A major feature of Flora is that it achieves evaluation performance comparable to frontier models at a low cost. The labeling cost per hour of video estimated in the paper is as follows: ・Flora: $0.72 ・GPT-6 Astra: $88.49 ・Claude Opus 5: $41.70 ・Gemini 3.8 Flash: $8.48 Under NVIDIA's evaluation conditions, Flora operates”
Exact source · revision 2Source owner not reported
Reported quantity · cost per hour of video: 41.7 other · Basis not reported · Differing source assertions
“criteria. These are the benchmark results reported by NVIDIA. 5. Processing cost and scalability A major feature of Flora is that it achieves evaluation performance comparable to frontier models at a low cost. The labeling cost per hour of video estimated in the paper is as follows: ・Flora: $0.72 ・GPT-6 Astra: $88.49 ・Claude Opus 5: $41.70 ・Gemini 3.8 Flash: $8.48 Under NVIDIA's evaluation conditions, Flora operates”
Exact source · revision 2Source owner not reported
Reported quantity · cost per hour of video: 8.48 other · Basis not reported · Differing source assertions
“criteria. These are the benchmark results reported by NVIDIA. 5. Processing cost and scalability A major feature of Flora is that it achieves evaluation performance comparable to frontier models at a low cost. The labeling cost per hour of video estimated in the paper is as follows: ・Flora: $0.72 ・GPT-6 Astra: $88.49 ・Claude Opus 5: $41.70 ・Gemini 3.8 Flash: $8.48 Under NVIDIA's evaluation conditions, Flora operates”
Exact source · revision 2Source owner not reported
Evidence dependency changes (8)
- 2026-10-30T22:02:36.872Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T22:02:21.211Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T21:50:09.499Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T21:49:49.733Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T21:39:15.610Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T21:39:11.977Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:49:41.756Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
- 2026-10-30T20:49:34.666Z · evidence updated · source revision 2. Evidence extraction was updated; current source attributions are shown above.
Report timeline
Follow attributed reports and material updates.
- Chinese robotics — Dataset and collection discoverySignalNVIDIA Opensources Robot Data Engine Flora, with Labeling Cost Just 1/125 of GPT-6 Astra
NVIDIA has released and open-sourced the robot data engine Flora, which transforms vast amounts of robot video data into high-quality training data using three models: action description, time segmentation, and reward models. Flora performs exceptionally well across multiple benchmarks, with labeling costs just 1/125 of GPT-6 Astra's, and can process 200,000 hours of video data per day.
- Chinese robotics — Dataset and collection discoverySignalNVIDIA Open Sources Robot Data Engine Flora, Annotation Cost is 1/125 of GPT-6 Astra
NVIDIA has released and open-sourced the robot data engine Flora, which transforms large volumes of robot video into high-quality training data through the collaboration of three models. Flora performs exceptionally well in multiple benchmark tests, achieving an annotation quality score of 57.1% and an estimated cost of $0.72 per hour of source video, which is only 1/125 of GPT-6 Astra's cost.
- Chinese robotics — Hardware and sensingSignalNVIDIA Expands Isaac Robot Platform with Humanoid Reference Platform Set for Early 2027
NVIDIA announced at GTC Berlin the expansion of its Isaac open robot development platform, including a humanoid robot reference platform, updated tools, and datasets. The Agile ONE SN, set for early 2027, integrates simulation, training, and deployment tools, with hardware and software designed for rapid development and real-time operation.
- Robotics — Robot learning news discoverySignalOverview of NVIDIA 'Flora' - A Suite of Open Models for Automatically Generating Robot Learning Data
NVIDIA has released 'Flora', a suite of open AI models designed to automatically generate robot learning data. Flora includes three models for action segmentation, captioning, and reward generation, and is claimed to outperform other models in certain benchmarks while being significantly more cost-effective for large-scale video processing.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).