H3-Image Day-0 Support: 4K in 0.87s with Vera Rubin

Developers brought the open-weight text-to-image and editing model MiniMax H3-Image to NVIDIA Vera Rubin using vLLM-Omni, achieving a 0.87-second median generation time for native 4K images on four GPUs. The system also reaches a throughput configuration of 6,115 images per hour, projecting substantial daily capacity and gross revenue potential under official API pricing examples. In addition to NVIDIA hardware, the model supports AMD, Intel, Huawei, and Moore Threads platforms. Open article →