TokenFactory
ModelsPrivate DeploymentGPU ClustersDocsBlog
Log in
Blog

Frontier Research

What the optimization loop found on real hardware, including the changes that did not pan out.

November 6, 20269 min read

Making MiniMax-H3 up to 9x faster on H100s

A 15-second clip went from 419 s to 46 s on four H100s. What worked, what only worked in the benchmark, and what we tried and closed.

MiniMax-H3VideoH100
Read →
TokenFactory
ModelsPrivate DeploymentGPU ClustersDocsBlogVideo ArenaPrivacyTermsData Policy
© 2026 Synvia Corp. All rights reserved.