Model Matrix#

Refer to the table below for world foundation models (WFMs) available in Cosmos Transfer2.5.

Model

Description

Minimum GPUs Required

Required GPU VRAM

Cosmos-Transfer2.5-2B

General checkpoints, trained from the ground up for Physical AI and robotics

1

65.4 GB

Cosmos-Transfer2.5-2B Auto

Specialized checkpoints, post-trained for autonomous vehicle (AV) applications

8

Inference Performance#

The following table shows generation times for different NVIDIA GPU hardware for single-GPU inference:

GPU Hardware

Cosmos-Transfer2-2B (Segmentation)

NVIDIA B200

285.83 sec

NVIDIA H100 NVL

719.4 sec

NVIDIA H100 PCIe

870.3 sec

NVIDIA H20

2326.6 sec

Note

Generation times are valid for video at 720P, 16FPS, 5 seconds length (93 frames), and with segmentation control input.