Egocentric video annotation: GPT-6 costs $100+/hour, Perceptron claims 8x cheaper
AkshatS07 · x · 2026-11-05
Perceptron launches its Egocentric annotation API, targeting ego-centric video as a pretraining-scale data source for video understanding and robotics. The bottleneck is labeling cost: GPT-6 runs >$100/hour, making 1M hours cost hundreds of millions. Perceptron's default 4fps setting scores 0.787 on its upcoming EgoHome benchmark—within 0.010 of GPT-6 Astra—at 1/8 the cost. The author claims to have labeled thousands of hours internally to build scaling laws for video understanding models.
More from Research
- De novo multistate interface design lets protein assemblies switch between geometries — DaveJuergens · 2026-11-05
- Stanford study: NVIDIA B200 delivers 4x the dollar value per unit of energy vs AMD MI350X — rohanpaul_ai · 2026-11-05
- ReuseGR decouples generative retrieval: 7B offline planner plus 0.5B online model cuts decoding cost — _reachsumit · 2026-11-05
- ZeroGR paper: filtering-first decoding speeds generative recommendation up to 9.9x — _reachsumit · 2026-11-05
- ByteDance's OneTrans-GSR unifies reranking and mix-ranking with one AR transformer — _reachsumit · 2026-11-05
- Scaling laws for data-constrained dense retrieval: multiplicative power law across BERT retrievers — _reachsumit · 2026-11-05