Hailuo 3.0 release date: MiniMax H3 launched 2026-07-31
Hailuo 3.0 is the name people use for MiniMax H3, which MiniMax released on 2026-07-31. What the vendor pages call it, and the ids that reach it on Sume.

"Hailuo 3.0" is the name people use for MiniMax H3, the video model MiniMax released on 2026-07-31. MiniMax's own pages call it MiniMax H3, not Hailuo 3.0: the Hailuo AI app page lists it under that name, and Sume lists it as minimax-h3.
Dates and specs below come from MiniMax's announcement, its model card, its API docs and the Hailuo page, all read on 2026-09-29. The Sume ids come from the Video generation docs.
Is Hailuo 3.0 the same model as MiniMax H3?
Hailuo is MiniMax's consumer video brand: the H3 model card links "Hailuo AI" to hailuoai.video and hailuoai.com as applications, and hailuoai.video hosts a page titled for MiniMax H3. That page describes a "Native Multimodal Video Model" for generation and editing from text, images, video and audio.
Treat "Hailuo 3.0" as a search phrase for that model. When you call an API, use the model id the provider lists, not the brand.
When was MiniMax H3 released?
MiniMax's announcement is dated 2026-07-31. It describes video with native stereo sound, up to 15 seconds at 2K, with a 768p option, and says MiniMax planned to open the model weights in the coming days. The model card on Hugging Face now exists, so the weights are published there; the announcement itself gives no later date.
| Page | What it says about H3 |
|---|---|
| MiniMax announcement | Dated 2026-07-31; up to 15 s at 2K, 768p also available, native stereo sound, native multi-shot |
| Model card (Hugging Face) | 33B-parameter dense Transformer; two checkpoints, FL2VA and Ref2VA; 4–15 s; aspect ratios 21:9 to 9:16 |
| Hailuo AI page | 5–15 s clips at 24 FPS, up to 2K, up to 12 reference files |
| MiniMax API docs | H3: 768P and 2K, 4–15 s; H3 Max: 480P and 768P, 5–15 s |
Can I call it through an API?
Yes. On Sume, MiniMax H3 is minimax-h3 and the faster variant is minimax-h3-max, both on POST /v1/videos. The docs list minimax-h3 at 5–15 seconds at native 480p and 768p, with 2K and 4K as priced upscales, and minimax-h3-max at 480p, 768p and 1080p. Note the vendor's own API docs list 4–15 seconds for H3; the Sume catalog range is 5–15.
Read supported_durations and supported_resolutions from GET /v1/videos/models before you submit, as the docs advise, since limits differ by model.
Is Hailuo 2.3 or an older model the same thing?
No. The pages read for this post describe H3 as its own model with its own limits. Sume's video catalog doc lists minimax-h3 and minimax-h3-max only for MiniMax; it does not list an older Hailuo model.
Sources
Related posts
More in Models
- How much VRAM for AI video generation? Numbers from model cards
From 14 GB to 80 GB: the VRAM figures Tencent, Wan-AI and Genmo publish for their open video models, and what each number was measured with.
- HunyuanImage 3.0 API: open weights, no listed Sume model
HunyuanImage 3.0 is an 80B-parameter open-weights image model from Tencent. Sume's image docs do not list it; here is what the card says and the hosted route.
- HunyuanVideo 1.5 low VRAM: 14 GB with offloading
Tencent's HunyuanVideo-1.5 card lists 14 GB as the minimum GPU memory with offloading on. The OOM settings and step-distilled option it names.
- HunyuanVideo API: is there a hosted one, or only weights?
HunyuanVideo is published as downloadable weights. Sume's docs list no HunyuanVideo API, so here is what running it takes and the listed video ids.
Written by Sume