MiniMax-H3 Is Now Publicly Available

About Licence: This regional carve-out stems from our ongoing generative video copyright litigation with major @Hollywood studios. We aim to deliver our AI tools globally to all users responsibly. US-based persons looking to deploy MiniMax-H3 may submit formal licensing requests. Licenses will be issued to applicants who commit to implementing robust compliance controls and guardrails fully aligned with US legal, compliance and regulatory requirements.

Amazing release!!!

Minimax-H3 finally arrived

Hope to see more video generation models like this to drive down the price of Seedance.

what minimax H3 is and how to use it

Thank you for making H3 open!! Let's run MiniMax H3 with SGL-Diffusion!

For the EU, UK, and South Korea. This regional carve-out is driven by ongoing regulatory evolution surrounding AI-generated video content across these regions. Unlike hosted services, open-weight model releases allow developers to independently deploy and modify model weights, introducing unique compliance challenges amid the current regulatory uncertainties in these regions. We will continuously monitor and assess evolving local regulations to adjust our access policies. In the meantime, persons based in these regions seeking to deploy MiniMax-H3 may submit formal licensing requests. Licenses will be granted to applicants that commit to implementing robust compliance controls and guardrails fully aligned with local legal and regulatory standards.

MiniMax-H3 is crazy good!

Love it

Online API Use MiniMax-H3 directly via API. Global: | CN: Online App Use MiniMax-H3 directly via App. WebApp Global: | CN: Desktop Global: | CN:

Congratulations 🔥

Detective mode: on. This MiniMax-H3 open weight drop just leveled up the whole case.

congrats! really cool release

running it

Congrats on this super release!

Do we need local deployment authorization from @MiniMax_AI to use this in the US?

We're genuinely thrilled to bring Hailuo 3 to Picsart creators! There's something special about putting a tool like this into the hands of people who will push it in directions even we haven't thought of yet! 💜🤩

@grok How much VRAM is needed to run this?

It's happening! 🥳🥳

Chinese TTV models be like: ‘Open weights, 2K, stereo audio, ready to cook.’ USA ones still stuck on ‘invite only’ and praying the waitlist blesses them 😂

going to test it right now

mini max h3 is out the only thing this model can maximize is the price of electricity

So awesome 🤩

This is huge!!! Thank you 🙏🏽

Awesome. Big thing for open weight community

Best time in human history by far and its only getting better Extremely Bullish on AI art a& entertainment even more than coding and SaaS A world of AGI still needs “bread and circus”

Hope to see more video generation models like this to drive down the price of Seedance.

Yaaay

OH?!

OMG, thank you!! 🤯

Let’s see how it goes for my next creative demo with MiniMax H3 Before:

We got supervised fine-tuning working on H3 🎉 ~150-line trainer + latent caching on top of the official Diffusers integration. Wrote up the conventions that silently break training. Feedback welcome:

这里有fine tunin的代码，欢迎指正。

Some points: Only 1 of the 3 modules is actually open (local = 768p), the encoder is Qwen3-VL-32B, and the released weights are CFG-distilled with sparse attention held back. Details