Elorian launches Glimpse multimodal model
Glimpse targets computer use, CAD and data labeling, but one user argues robotics is its biggest use case, with per-step camera checks that are 7x cheaper than GPT-6 Astra.
Elorian launched Glimpse today. Most of the talk is about computer use, CAD, and multimodal data labeling. I think robotics will be the biggest use case. Robots can already move. What they’re missing is a planner that sees well enough to tell whether a move worked, and is fast and cheap enough to check every step. Demo below: Glimpse plans each step, then checks the cameras every 2s to confirm it actually happened. A check every 2s is 1,800 calls per robot-hour. Glimpse is 7x cheaper than GPT-6 Astra and ~1s faster per call. The same budget on Astra buys one check every 14s. Computer use, CAD, and labeling are the near-term revenue. Robots come later and get bigger. Today, Glimpse can tell a robot a step failed so it retries. Next, a robot that retries can learn on the job, graded on what actually happened. I rebuilt it as a simulation where your hand is the robot arm. Same camera, same executor, two planners. Try it: http://brianzhan.com/glimpse
This post quotes another post: x.com/andrewdai/status/2111142161584628060