MiniMax H3: the open-weight omni model

What changed in MiniMax H3

Versus prior Hailuo and MiniMax video models, H3 is the first built as a single omni transformer merging text, image, video and audio, the first at native 2K, and the first with open-sourced base weights. On 27 August 2026 the inference platform fal launched H3 Max, a post-trained version of the open H3 weights built with input from the MiniMax H3 team: it renders a 5-second 768p clip in about 3 seconds, at 480p or 768p and 5 to 15 seconds per clip. H3 Max pricing is not verified from more than one source.

How to get MiniMax H3

The Hailuo app and platform, the MiniMax Open Platform API, and third parties including OpenRouter, Morphic and EvoLink. Self-hosting only where the license permits.

What MiniMax H3 means for short-form marketers

The open-weight base model is the real differentiator for teams with technical resources: self-hosting or fine-tuning on brand assets removes per-second API costs, and single-pass audio plus video with up to 12 references suits fast, localised ad variants where you swap the voice, face or product reference and regenerate.

The catch is geography. The license explicitly excludes the US, UK, EU and South Korea, so most Western marketing teams will use the paid hosted API, where native 2K with heavy reference inputs is slower and pricier per generation than simpler competitors. It is also newer and less proven in production ad pipelines than Kling or Veo.