We make expensive computation cheap enough to ship.
For four years that meant zero-knowledge proofs and the ASICs that run them.
Today it means video. Phosor, our generation platform, renders a
second of 480p video for half a US cent — text to video, image to video, lip-sync,
motion transfer, your own LoRAs, and an API with metered billing behind all of it.
16 frames = one second of generated video. At 480p turbo, that second costs $0.005.
$0.005per second, 480p
1080presolution ceiling
11model capabilities
$0.003per LoRA train step
Pay as you gono seat licence
Published rates, read from the platform's own pricing endpoint on 22 September 2026.
Per-second figures are the published per-frame price at 16 fps.
Full table →
Two lines of work
One problem, two decades of hardware.
Proof generation and video generation are the same engineering problem wearing
different clothes: an enormous arithmetic workload, a hard latency budget, and a
cost per unit of output that decides whether anyone can afford to use it.
Live business
Video AI
Open-weight video models — Wan 2.2, LTX, MiniMax H3 — served at metered prices,
with LoRA training, e-commerce imaging and a short-drama workflow on top.
It runs as Phosor, a platform you can sign up for today.
Text, image, speech and multi-reference driven generation
Custom silicon and software co-design for zero-knowledge proving: ZKVM
accelerators reaching more than 20x a GPU prover on the same
workload, delivered with partners into production mining hardware.
Hardware/software co-design, from the arithmetic up
Crypto and ZKP IP libraries shipped in tens of millions of chips
Eleven published papers on RISC0, SP1, Binius, Jolt and Aleo
The gap between open weights and the closed frontier is mostly post-training and data curation, not architecture. Here is the evidence, and what it costs to close.
Open weights reached the top of the human-preference leaderboard in August 2026. What that changes, and the three constraints that stop it being a clean win.