All models

MiniMax H3 on VM0. Native-audio video up to 2K

MiniMax's open general-purpose video model. Generate 4 to 15 second clips from text, images, video and audio with native stereo sound.

Video / Text-to-video / Image-to-video / Audio

MiniMax H3 is an open, general-purpose multimodal video model. It understands text, images, video and audio in one context, then generates 4 to 15 second clips at 768p or 2K with native stereo sound. On VM0, the model accepts up to 9 reference images, 3 reference videos and 3 reference audio clips, with a maximum of 12 reference files in one request.

H3's clearest position in the curated lineup is high-resolution, reference-guided video with audio. It reaches 2K while handling motion, character, product and soundtrack references together. Dreamina Seedance 2.5 is the better fit for 30-second stories and larger reference sets; Veo 3.1 Fast and Kling V3 4K remain the 4K options.

What is MiniMax H3?

August 2026 · Open general-purpose multimodal video model with native joint audio and video generation.

MiniMax introduced H3 in July 2026 and open-sourced the model in August. The release combines unified text, image, video and audio understanding with native 32 kHz stereo sound, output up to 2K and clips up to 15 seconds.

VM0 exposes the hosted MiniMax route through the minimax-h3 alias. It supports text-to-video, first-frame, last-frame and first-to-last-frame generation, plus a separate reference mode for images, video and audio. Boundary frames and reference media cannot be combined in the same request.

What's notable about MiniMax H3

Headline architecture and capability features.

General-purpose multimodal video generation system with joint video and stereo-audio output. H3 unifies text, image, video and audio context for reference-guided generation. Its 2K path regenerates from the 768p result with the original context to preserve detail at the higher resolution.

Specs at a glance

FamilyMiniMax H3
ModalitiesText, image, video and audio to video
Clip durations4 to 15 seconds
Output resolutions768p / 2K
Output audioNative 32 kHz stereo
Reference inputsUp to 9 images / 3 videos / 3 audio clips; 12 files total
Frame controlsFirst frame and last frame
Prompt limit7,000 characters
VM0 usage price$0.10 / second at 768p; $0.163 / second at 2K
Available on VM0August 2026

MiniMax H3 pricing

Vendor list price per generated unit.

Per second of video$0.10
Detail768p output; 2K is $0.163 per second

How MiniMax H3 behaves in practice

Observed behaviour from production agent runs.

2K output

The highest resolution reference-guided option in the curated lineup. H3 can take image, video and audio references all the way to 2K instead of stopping at 720p or 1080p.

Native stereo audio

Every H3 generation includes jointly generated stereo sound. Voice, effects and music can develop with the visual sequence instead of being added in a separate pass.

Omni-reference control

A single brief can combine characters and products from images, motion from videos, and voice or rhythm from audio. H3 is especially useful when the desired output depends on relationships across those reference types.

Reference cost visibility

Output and input video seconds are billed at the same resolution-specific rate. The first five reference images are included; each additional image is $0.05 on VM0. Reference audio is not billed separately.

Best agent tasks for MiniMax H3

The 2K product film with sound

Use product and brand images for visual control, add a motion reference and soundtrack, and generate a high-resolution clip with native stereo audio in one pass.

The performance guided by voice and motion

Combine a character image, a movement reference and a voice clip. H3 can use all three modalities together to coordinate appearance, action and sound.

The first-to-last-frame reveal

Set the opening and closing frames and let H3 generate the transition between them at 768p or 2K with native audio. This mode is separate from multimodal reference generation.

When to skip MiniMax H3

Skip MiniMax H3 when you need a clip longer than 15 seconds, a silent output, more than 12 reference files, or boundary frames and reference media in the same request. Use Dreamina Seedance 2.5 for 30-second narratives and larger reference sets, Veo 3.1 Fast for short 4K cinematic output, or Kling V3 4K for stylised native 4K video.

MiniMax H3 vs other models

MiniMax H3 vs Dreamina Seedance 2.5

MiniMax H3 leads on output resolution with 2K video and always-on native stereo sound. Seedance 2.5 leads on duration and reference capacity, with clips up to 30 seconds and as many as 50 multimodal references.

MiniMax H3 vs Veo 3.1 Fast

H3 generates longer clips up to 15 seconds and accepts image, video and audio references, while Veo 3.1 Fast tops out at 8 seconds but reaches 4K. Both generate native audio; choose H3 for reference control and Veo for the 4K cinematic finish.

MiniMax H3 vs Kling V3 4K

H3 combines multimodal references, native stereo audio and output up to 2K. Kling V3 4K reaches a higher delivery resolution and has a stronger stylised ceiling, but does not offer H3's reference-guided workflow.

Bottom line: should you use MiniMax H3?

Choose MiniMax H3 for reference-guided clips that need native stereo audio and resolution up to 2K. Choose Seedance 2.5 for longer narratives and larger reference sets, or a 4K route when final delivery resolution is the priority.

Frequently asked questions

What duration and resolution does MiniMax H3 support on VM0?

Every whole-second duration from 4 to 15 seconds, at 768p or 2K. The current VM0 route defaults to 2K.

Does MiniMax H3 always generate audio?

Yes. H3 always returns native stereo audio with the video. A silent-output option is not available for this model.

How many references can one H3 request include?

Up to 9 images, 3 video clips and 3 audio clips, with no more than 12 reference files in total. Reference audio must be accompanied by at least one image or video.

Can first and last frames be combined with other references?

No. H3's boundary-frame mode and multimodal reference mode are separate. A request can use first and last frames or reference images, videos and audio, but not both groups together.

How is MiniMax H3 priced on VM0?

Output costs $0.10 per second at 768p or $0.163 per second at 2K. Reference-video duration is billed at the same rate. The first five reference images are included, then each additional image costs $0.05; reference audio is free.

Is MiniMax H3 open source?

MiniMax released H3's task-specific checkpoints and supporting code in August 2026. VM0 uses MiniMax's hosted API route, so you can run the model without deploying those checkpoints yourself.

Alternatives

Using MiniMax H3 on VM0

Using MiniMax H3 on VM0

VM0 agents can call MiniMax H3 as part of an agent run, billed against your VM0 credits. The list price above is what the upstream provider charges; VM0 passes that through with the standard credit conversion.

Available on VM0 since August 2026.