🦾TwelveLabs Pegasus 1.6 Labels First-Person Robot Video
TL;DR
TwelveLabs released Pegasus 1.6, tuned for first-person footage used to train robots. In internal testing it processed roughly 17 years of video in about 18 hours with no failed files.
TwelveLabs released Pegasus 1.6, tuned for first-person footage used to train robots. In internal testing it processed roughly 17 years of video in about 18 hours with no failed files. It describes scenes; it does not map them to motor control.
Key Points
Pricing: $1.75 per video hour of input, $3 per million input-image tokens, $7.50 per million output tokens
Each segment definition in a Segment request is billed separately, so multi-category annotation multiplies cost
CEO Jae Lee says Pegasus supplies descriptions and structure; developers still link observations to actions
Competes with Nvidia Cosmos Curator, Encord and Google Gemini Robotics ER 2
Why It Matters
Robot data pipelines are the bottleneck, and video labeling is where the money is moving. Watch the per-segment billing before you scale.
Quick Facts
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,556 builders reading daily.