pi-H3: Open-Source Model Generates Videos in Seconds on B200

Stanford's Gordon Wetzstein team and Luma AI open-sourced pi-H3, a distilled MiniMax H3 that generates a 5-second video in 3.8 seconds on a single B200 by using only 2-4 sampling steps and skipping the VAE decoder.

2026-11-11 ~ 2026-11-11 · 2 related posts

• pi-H3 generates 768p video in 2-4 steps: 10s clip in 9.2s on a single B200 — GordonWetzstein · 2026-11-11
• pi-H3 generates video in 2–4 steps with no VAE decoder — 5s clip in 3.8s on one B200 — Apprehensive_Sky892 · 2026-11-11