If you've been seeing "Hailuo 3" in AI newsletters, Reddit threads, or video demo posts over the last week, the search that usually follows is "MiniMax H3 release date." The short answer: MiniMax officially announced MiniMax H3 on July 31, 2026, through its launch post and model release notes.
That announcement matters because H3 is not just a version bump. MiniMax describes it as an open, general-purpose, omni-modal video model — text, image, video, and audio in; native audio out — and it is the model behind the clips people loosely call "Hailuo 3." This explainer is current as of August 2026 and built from MiniMax's own pages, so every date and claim below is traceable to a first-party source. By the end you'll know the verified release date, what actually changed, what is still likely to change, and what to verify before you pay for credits or build an integration.
Pain This Article Solves
A "release date" search is rarely about a date alone. The question underneath is usually: Is MiniMax H3 actually available, what can it do, and should I start now or wait? Most release-date pages only answer the first half. This one also proves the date with official links, explains what changed versus older Hailuo workflows, flags what is still moving, and gives you a checklist to run before production.
Official Release Date
MiniMax published the MiniMax H3 launch post and the H3 release notes on July 31, 2026. That is the official, traceable date — most other coverage online is a derivative of it.
What the release notes say about H3:
- Open, general-purpose, omni-modal video model
- Inputs: text prompts, image references, video references, and audio
- Native audio generated as part of the output
- Up to 2K resolution, roughly 4–15 seconds, 24 fps
- Runs through an async API workflow on the MiniMax platform
- Open weights expected "in the coming days" — note there is no fixed date
- A technical report is "coming soon," also without a date
Two things are easy to miss here. First, "Hailuo 3" is informal shorthand: Hailuo is MiniMax's consumer product brand, and MiniMax H3 is the model name behind it. Second, MiniMax positions H3 as a model family spanning video, voice, and music generation, so the same release touches more than just video pipelines.
MiniMax H3 Timeline
| Date | Event | Why it matters |
|---|---|---|
| July 31, 2026 | MiniMax H3 launch post and release notes published | The official, traceable release date |
| August 2026 | This guide re-checked official docs | Current snapshot; some pages still lag |
| "In the coming days" | Open weights availability (per release notes, no fixed date) | Treat any promised download date as unconfirmed |
| Not announced | H3 video pricing and technical report | No official numbers exist yet — do not quote any |
This timeline deliberately excludes rumored dates. If a date cannot be traced to MiniMax or a first-party page, it is not treated as authoritative here.
What Changed With MiniMax H3
The headline change is not "a newer model exists." The change is the input and output surface:
- Text prompts still drive scene direction
- Image references now control subject, style, and composition
- Video references steer motion and visual pacing
- Audio joins the input set, and H3 generates native audio instead of requiring a separate voice track
- Output stays short-form: roughly 4–15 seconds at up to 2K and 24 fps
Practically, that changes what you can brief. A marketer can start from a product photo. A filmmaker can feed a reference clip to lock a camera move. A social creator can test audio-aware scenes without stitching audio in afterward.
A quick technical-depth note on the workflow
H3 runs through an async API: you submit a generation task, receive a task ID, poll the task until it completes, then retrieve the result. That is the same create-and-query pattern MiniMax documents for video generation, and it matters for integrations — a queued request is normal behavior, not a failure.
Rule of thumb: if your target output is under about 15 seconds and you need reference control or native audio, H3 is worth testing now. If you need long-form video or a fixed per-second price, wait — those still aren't announced.
Should You Start Using MiniMax H3 Now?
Yes — but with small tests first. H3 is live through MiniMax's platform, so you can validate quality today without betting a production pipeline on it.
| Your situation | Move |
|---|---|
| Need short-form AI video with reference control or audio | Start testing H3 now on small drafts |
| Need long-form video | Wait — H3 is short-form only right now |
| Need to quote a budget or price H3 into a product | Wait — video packages don't support H3 yet, so there is no official price to quote |
| Building an API integration | Start against the async create/query endpoints, but pin model name and limits to current docs |
| Waiting on open weights | Watch the release notes — "in the coming days" is a commitment, not a date |
A low-risk first test: open MiniMax H3 AI, feed it one product or character image, write a single-shot prompt, and generate a short draft. Review subject consistency, motion, camera, and audio before you scale anything up. And when you're ready, try the H3 AI video generator directly — it's the fastest way to see whether the new input surface actually helps your workflow.
What to Verify Before Production
AI model documentation changes faster than typical SaaS feature pages. Before you spend budget or ship a feature, verify:
- Current model name and availability in the release notes
- Supported input modes for the task you're building
- Duration, resolution, and fps options actually served
- Pricing per second and whether pay-as-you-go covers H3
- Whether video packages support H3 yet (they do not, as of this writing)
- Rate limits and concurrency for your usage pattern
- Result retrieval and storage behavior in the async workflow
- Content and rights policies for commercial use
If any page contradicts this article, the official docs win.
FAQ
When was MiniMax H3 released?
MiniMax announced MiniMax H3 on July 31, 2026, via the official launch post and release notes.
Is MiniMax H3 available through an API?
Yes. MiniMax's platform docs list MiniMax-H3 in the video generation family with an async create-and-query workflow. Confirm the current model name and limits in the docs before integrating.
Is MiniMax H3 the same as Hailuo 3?
Not exactly. H3 is the model name; Hailuo is MiniMax's consumer product brand. "Hailuo 3" is informal shorthand people use for videos generated with H3, but official pages use MiniMax H3.
Should I trust non-official release-date pages?
Only for discovery. For dates, specs, pricing, and API behavior, rely on MiniMax's release notes, launch post, and documentation — that is what this article does.
Should I wait for the open-weights release?
That depends on your goal. If you need self-hosted or offline inference, the release notes promise open weights "in the coming days," but no date is fixed. For hosted generation, there is no reason to wait — the API is live now.
Want to skip the checklist anxiety? Open the MiniMax H3 AI video generator and run a single test clip — that is the fastest way to confirm whether H3's inputs and native audio fit your next project. The generator wraps MiniMax H3, so you are testing the real model, not a demo.
Sources
- MiniMax H3 launch post — Official July 31, 2026 launch announcement.
- MiniMax H3 release notes — Official model release reference, including open weights and technical report notes.
- MiniMax video generation guide — Official video workflow context.
- MiniMax models introduction — Official model capability reference (resolution, duration, fps).
- MiniMax create video generation task API — Official async task creation endpoint.
- MiniMax query video generation task API — Official task polling and retrieval endpoint.
- MiniMax pay-as-you-go pricing — Official pricing reference; confirm H3 is covered at current rates.
- MiniMax video packages — Official package support reference; H3 not yet listed.
- MiniMax rate limits — Official limits reference.
- MiniMax homepage — Official company source.
