Blog
News and perspectives from the team at Reactor
From 49 Steps to Real Time: Inside FastH3-Omni
Inside FastH3-Omni: the architecture, math and engineering behind real-time H3.
OCT 21, 2026
Optimizing the Wan VAE decoder for real-time world models
We cut the Wan VAE decoder's time per chunk from 133.8 ms to 60.2 ms, 2.2× faster, at the same output quality.
OCT 12, 2026
Reactor Expands Investors in Series A Round to Scale Real-Time World Model Infrastructure
Investment from NVIDIA and Sapphire Ventures comes as Reactor sees growing adoption across media, entertainment and physical AI
OCT 5, 2026
A kernel-centric path to real-time video generation on Trainium
Using the Neuron Kernel Interface, a Reactor–AWS collaboration tackled the dynamic shapes, memory access patterns, and cache management that make real-time autoregressive diffusion hard—building techniques that generalize across models.
SEP 25, 2026
Announcing Sol-H3 on Reactor: even faster MiniMax-H3, built by NVIDIA
Reactor partners with NVIDIA's SANA team to launch Sol-H3, powered by Sol-Engine: 5 seconds of 1344×768 video with native audio in 1.67 seconds on Blackwell, 3× faster than real time.
SEP 7, 2026
Visko Orbis 1.0 Launches on Reactor
Reactor partners with Visko to exclusively launch Orbis 1.0 — an infinite-length realtime video generation model, in Stable and Dynamic variants — on Reactor starting day zero.
SEP 1, 2026
LTX is Now Available on Reactor
Reactor partners with LTX to bring their newest open world video model — realtime, high-resolution, with native synchronized audio — to Reactor on day zero.
AUG 10, 2026
How to train a frontier-level world model
We are releasing and open-sourcing the model and training code, and sharing all the lessons we learned along the way in this blog post.
JUL 21, 2026
Introducing Reactor: The Developer Platform for World Models
Reactor emerges from stealth with $59M to build the platform for real-time World Models
MAY 28, 2026