
HyperFlow H3
A data-free flow self-distillation technology built on MiniMax H3.
- Lower inference cost
- Frontier model quality
- Open source
Ranked #2 on Artificial Analysis Video Editing v1.1 leaderboard, Nov 2026.

Introducing HyperFlow H3
See it in action
Showing Camera Control
How it works
The model generates its own candidate videos. Filters them for quality. Self-distills from the ones that pass.
No external data. No human labels. The model trains itself with its own outputs.
This is the self-distillation component of RSI (Recursive Self-Improvement), our training paradigm for video models.
HyperFlow H3 also supports NVIDIA's Sol-Attn sparse-attention kernel, delivering up to 13% faster H200 generation than dense attention.
The HyperFlow series
The series now includes three releases, each addressing a different constraint. HyperFlow reduced the cost of training. HyperFlow Infinity extended the length of a single generation. HyperFlow H3 Flash reduces latency until generation and playback run at the same speed.


A recursive agent harness for infinite video generation.
View InfinityInfinity and HyperFlow share the same weights. One download works for both.
HyperFlow H3 Flash
Real-time generation at 768p on four B200 GPUs, and the same weights on consumer hardware.








