SambaNova has launched prompt caching for MiniMax M3 on SambaCloud, reducing input costs by 90% to $0.06 per million cached tokens and improving time to first token by 35% to 88% without code changes.
Hao AI Lab announced that FastVideo FastH3 runs on single consumer machines, including NVIDIA RTX GPUs and Apple Silicon, and released the compact FastH3 Trim model requiring 8 GB of memory.
Replicate is hosting Pruna AI’s P-Video-2 Pro and places it on the preference-versus-speed and preference-versus-price Pareto frontiers. Design Arena says the Quality and Speed variants tie for second on its image-to-video leaderboard.
In ComfyUI, encoding is up to about 2.2 times faster and decoding is about 1.4–2.7 times faster. ComfyUI reports a 1344×768, 129-frame round trip falling from 24.3 seconds to 12.7 seconds.
Perplexity said Perplexity Computer can now create videos using MiniMax H3 and ByteDance Seedance 2.5, producing finished video next to copy and creative in the same thread. It said the capability is available now for Pro and Max subscribers.
ComfyUI said it generated a shot in less than a minute with MiniMax H3 Max via Partner Nodes. It described the workflow as useful for fast video-to-video experiments from a basic blockout plus reference images, and said the workflow includes an option to upscale to 2K.
