The Developer Platform for World Models.
San Francisco
Joined November 2025
- Tweets161
- Following1
- Followers9.7K
- Likes438
Introducing NVIDIA Cosmos-3-Nano on Reactor. Two new models, available now via the Reactor API.
• Sim-Bimanual predicts what a robot sees as you move its arms and grippers.
• Sim-Transfer turns simulations into photorealistic video.
Try them now: reactor.inc/cosmos-bimanual
reactor retweeted
Introducing NVIDIA Cosmos-3-Nano on Reactor. Two new models, available now via the Reactor API.
• Sim-Bimanual predicts what a robot sees as you move its arms and grippers.
• Sim-Transfer turns simulations into photorealistic video.
Try them now: reactor.inc/cosmos-bimanual
Sim-Transfer: turn simulation into photorealistic video.
• Create training examples across lighting and material conditions.
• Re-render robot demonstrations in new environments.
• Turn rough 3D scenes into realistic previews.
docs.reactor.inc/robotics/co…
Sim-Bimanual: preview robot actions before running them on hardware.
• Compare predicted outcomes of action plans.
• Prototype robot controls without physical hardware.
• Generate robot’s-eye demonstrations from action sequences.
Try it here: reactor.inc/cosmos-bimanual
Try the model here: reactor.inc/models/eden-live
Excited to share the official open source release of FastH3 Omni in collaboration with @haoailab, following last week’s preview.
Try the model now on Reactor: reactor.inc/models/fast-h3-o…
As promised, FastVideo open-sources @MiniMax_AI FastH3 Omni, training code included! 🚀
- Real-time gen @ 12x faster on NVIDIA Blackwell
- NVFP4/FP8 for Spark + RTX!
- ComfyUI day-0
Collab with @NVIDIAAI , Nuva Lab @nuvalab and Reactor @reactorworld
Open source must win! ♥️
Join us next week at Long Horizon Summit by @encord_team and learn about robotics on Reactor!
📅 Oct 20–21
📍 Fort Mason, San Francisco
🤖 Booth SU #1
There may or may not be limited edition Reactor swag if you track us down...
Join us in conversation with Will Lin from FastVideo @haoailab as we get technical about all things realtime video!
Live from 12:30 PT.
xitter.cf/i/broadcasts/1jxXgBlMA…
The future of media is so bright
In celebration of our FastH3-Omni launch, I vibecoded a pre-visualization studio. Now today, I'm down in LA showing it to studios.
What you're seeing here is generated all in real-time, just from reference images and prompts. ⚡️
Open Source repo is in the comments to get started with this on the Reactor API!
We're honored to be named to the #Disruptors60 Class of 2026 curated by @Greenfield Partners at @TechCrunch Disrupt.
Thank you to our team, and to the customers and partners who helped us get here.
reactor retweeted
Introducing FastH3-Omni.
The fastest and highest-quality open-source version of realtime H3 by @MiniMax_AI, post-trained by FastVideo @haoailab with @nuvalab.
Exclusive preview on the Reactor API until 10.20.2026.
• First and last frame conditioning
• Audio input + strong dialogue
• Up to 3 reference images
• Frontier-level performance
Introducing FastH3-Omni.
The fastest and highest-quality open-source version of realtime H3 by @MiniMax_AI, post-trained by FastVideo @haoailab with @nuvalab.
Exclusive preview on the Reactor API until 10.20.2026.
• First and last frame conditioning
• Audio input + strong dialogue
• Up to 3 reference images
• Frontier-level performance
We made the Wan VAE decoder 2.2x faster for real-time world models.
Decode time per chunk went from 133.8 ms to 60.2 ms with the same output quality, 10% faster than vLLM-Omni’s fast path and 1.7x faster than SGLang’s.
Wan-based decoders are used in Cosmos 3, DreamZero, Lingbot-World 2 and many other world models. On a single B200, Lingbot-World goes from about 36 to 46 fps.
Read the full article: reactor.inc/blog/optimizing-…
