Launching today
Perceptron Mk1
Frontier video reasoning for the physical world
9 followers
Frontier video reasoning for the physical world
9 followers
Perceptron Mk1 brings frontier video and embodied reasoning to production APIs, with temporal grounding, structured visual outputs, 32K multimodal context, and pricing built for high-volume physical-world tasks.
Flowtica Scribe
Hi everyone!
Perceptron Isaac was their strong open-source series. Mk1 (Mark One) is the closed-source flagship — a production-oriented world-understanding model built for the physical world.
It delivers frontier-level video and embodied reasoning: accurate temporal grounding, multi-view understanding, pixel-precise pointing, reliable counting in dense scenes, and strong structured document extraction.
All while running at a much more practical cost than larger frontier models.
Fere AI
This is very cool. This currently works over a video. I am wondering if someone can make an almost real time video analysis using this? Maybe have a streaming video captured into smaller videos -> sent -> analysed -> stored in memory and then repeat until the video is going. What kind of videos are harder to process vs easy use cases?