Skip to content

Model catalogMiniMax

MiniMax H3

Omni-modal video + stereo audio generation — the open-weights Hailuo line. Served with vLLM-Omni behind your own private, video-generation /v1 endpoint — billed by the second, terminate anytime.

33B dense omni · bf16 · 4× 141 GB · per-second billing

01Specifications

The specs the launcher enforces.

Architecture, served precision, max context, the minimum GPU shape, and the exact pinned weights — the facts that decide how MiniMax H3 runs and what it costs.

SpecValue
Architecture33B dense omni
Served precisionbf16
Output4–15 s video @ 24 FPS + 32 kHz stereo audio
Minimum GPU shape4× 141 GB
Weights servedMiniMaxAI/MiniMax-H3@42ed227

Runs on 4× 141 GB · billed per second — final cost depends on the GPU shape.

Full pricing →

02How it runs

A private endpoint, not a shared API.

Launching MiniMax H3 runs the model on your own instance and gives you a private /v1 video endpoint and API key. Post a prompt (plus optional reference images, clips, or audio) as multipart form data to /v1/videos/sync and get back an MP4 with stereo audio; /v1/videos is the async job-polling variant. The endpoint is yours alone — terminate the instance and it goes away with it.

Billing is by the second on the 4× 141 GB shape the launcher enforces, from boot to terminate. No commitment and no tier gates.

03Get started

Launch MiniMax H3 now.

One private video endpoint, one key, billed by the second. The console opens with MiniMax H3 preselected.