Nvidia open-sources Flora, 125x cheaper than GPT-6 Astra for robot video labeling

nvidia open-sourced flora, a data engine designed to label robot videos at a significantly lower cost than gpt-6 astra. according to nvidia, flora can label one hour of video for 72 cents, while gpt-6 astra costs $88.50 per hour. flora covers three types of labels: action descriptions, step-by-step ratings, and task completion judgments. it uses three fine-tuned qwen3.5 models, including a 9b model for video segmentation and two 27b models for description and rating. in tests, flora scored slightly better than gpt-6 astra, with 57.1% annotation quality at $0.72 per video hour, compared to gpt-6 astra's 53.7% at $88.49. flora is already being used to train nvidia's gr00t models, annotating up to 200,000 hours of video per day across 512 gb300 gpus.