Sol-H3 team reports generating five seconds of AI video in 1.653 seconds

The team announcing Sol-H3, a MiniMax-H3 inference stack, says it generated five seconds of 1344×768 video at 24 fps with stereo audio in 1.653 seconds on eight Nvidia B300 GPUs. It reports an 11.04× speedup over Base H3 Dense for that clip length on the same hardware. The benchmark compares four-step Sol-H3 with 50-step Base H3 Dense—a full-profile comparison, not just an attention-only change. The team says figures are medians of three measured runs after one warmup. Timing includes text encoding, denoising and video/audio decoding, but excludes model loading, compilation warmup and final MP4 encoding.