Pricing

MiniMax H3 — the official name for what the community calls Hailuo 3 or Hailuo 3.0 — is MiniMax's third-generation omni-modal video model, launched July 31, 2026. It reads text, images, video, and audio as one unified context and generates clips at native 2K (1440p short edge), 24 fps, with stereo dialogue, sound effects, and ambience in a single pass. H3 Video is the online workspace that routes your prompts to compatible H3 backends — no local install.

MiniMax positions H3 as an open-weight model and plans to release checkpoints under the MiniMax Community License — free non-commercial use, and commercial use for organizations under $20M annual revenue with attribution. Weights were promised in the days after the July 31 launch, but no Hugging Face repo or GitHub release was public yet as of early August 2026. Until then, access is via API and hosted platforms like H3 Video.

Not on H3 Video. ComfyUI added an official MiniMax partner node (API-based) on launch day; local ComfyUI workflows will need the open weights once MiniMax publishes them. H3 Video skips all of that — write a prompt, upload references, and generate in the browser while the community waits for HuggingFace and GitHub releases.

Output: native 2K (1440-pixel short edge, ~3.7 MP at 21:9), 24 fps, 4–15 second clips (integer lengths), seven aspect ratios plus auto. Inputs: up to 9 reference images, 3 video clips (≤15 s total), and 3 audio files per generation — omni-reference for locking character, motion, and voice. Prompts up to 7,000 characters. MiniMax has not published parameter count; the official blog notes model size as a future scaling target.

On Artificial Analysis, H3 ranks #1 in video editing — ahead of Seedance 2.0 — while trailing Google Gemini Omni Flash in text-to-video and sitting behind both Seedance 2.0 and Gemini in image-to-video. MiniMax claims 2K per-second pricing under one-third of mainstream models (~$0.13/s reported). H3 Video lets you test these capabilities without API setup.

MiniMax previewed H3 at WAIC 2026 (July 17) and launched it July 31, 2026 on the MiniMax Open Platform API and Hailuo AI consumer app. Third-party hosts (fal.ai, OpenRouter, etc.) added support the same week. Open weights for local/HuggingFace use are promised but not yet published as of early August 2026.

Yes — new accounts get free credits, no credit card required. Each generation consumes credits based on duration and resolution; the cost is shown before you render. MiniMax's own API lists 2K at roughly $0.13/second — H3 Video converts that into simple credit packs.

Yes. Reference images and generated results are private to your account. See our Privacy Policy for data handling details.