TRACKEDOFFICIAL CLAIM
MiniMax H3
BY MINIMAX
The open general-purpose omni-modal video model underneath the H3 Max family.
Open MiniMax open-source announcementREALTIME VERDICT
NOT YET CLASSIFIEDThe capability is tracked, but no comparable RTR value is currently published.
LAST VERIFIED · 2026-09-03BEST RTR—No comparable result
OUTPUTUp to 2KFPS not reported
DURATION4–15 secondsAPI / local deployment
ACCESSAPIPrice not reported
WHAT WE KNOW
Capabilities with evidence attached.
01
Open weights and inference code are available.
02
Supports text, image, video, and audio context.
03
Tracked as the baseline for fal's faster post-trained variants.
BEST FIT
Video generationModel researchCustom inference
EVIDENCE GAP
Can you reproduce this model?
We are looking for endpoint, region, queue, first-frame, and last-byte timing across repeated runs. Raw JSON is more valuable than a screenshot.
Contribute a run