Kyutai CADRE: Video Autoencoders That Cut World Model Lag Below 230 ms

Kyutai has released CADRE, open causal diffusion representation autoencoders for interactive world models. The Kyutai CADRE video autoencoder targets one core problem: action lag. Many world models borrow autoencoders from text-to-video generators. Those autoencoders pack many frames into a single latent, so every action must wait. According to the research paper, CADRE world models run at 20 fps on a single H100 and respond within 230 ms.

TL;DR

• Size: Seven autoencoders, each with roughly 476M trained parameters plus a frozen 86M DINOv3 ViT-B/16. The world model has 1.2B parameters. Video runs at 288×512, 20 fps.
• Runs on: One H100 (bf16, batch 1). Quantized variants: not disclosed.
• Performance: CADRE defines most of the latency-quality frontier. It is as fast and better in 42 of 48 comparisons with MIRA, Wan 2.2, Cosmos, and LTX-2.
• Best: A driving FVD of 162.6 (CADRE-2×16×16, 1 step) versus 599.7 for Wan 2.2.
• Worst: Robot manipulation action following (ARR) of 0.45 for CADRE-2×32×32 versus 0.85 for Wan 2.2.

What Is CADRE?

(Original article content truncated — add the remaining paper summary here.)

via MarkTechPost