TensorScale Unveils Fastest Video Inference, Claims 10x Speedup for MiniMax H3

Scobleizer · x · 2026-08-12

TensorScale comes out of stealth, claiming the fastest video inference in the world. Their stack runs MiniMax H3, a leading open video model, 10x faster and at half the cost. Founded by Yao Zhang and colleagues, they aim to rebuild AI infrastructure for video and world models.

Original post →

More from Infra

Infra channel →