A Frontier AI Model That Anyone Can Host
MiniMax M3 matches the best closed AI models on coding tasks — and costs twelve times less. Soon, you'll be able to run it yourself.
What just happened
On June 1st, a Chinese AI lab called MiniMax released a new model — M3 — that's turning some heads. Not because it's flashy, but because of what it manages to do at the same time: handle incredibly long conversations, read images and video, write code as well as the best paid tools, and do all of this at a price that's about twelve times cheaper than Claude Opus or GPT-5.5.
What makes that last part interesting is that, around June 11th, they're planning to release it openly — meaning anyone with the right hardware could run it on their own servers, without paying per use.
Why this matters for your business
If you've ever used an AI tool and winced a little when you saw the bill, this is the direction things are heading. Models that were only available from big, expensive platforms are starting to appear as open alternatives.
For a restaurant group tracking supplier conversations across months of emails, or a design agency feeding an AI a full project archive to generate consistent copy — the ability to process enormous amounts of information affordably starts to become genuinely useful, not just a nice idea.
The caveats are real: this model comes from China, which means its API usage falls under Chinese data privacy law. Independent tests haven't confirmed all the claims yet. But the trend it represents — frontier capability at open-source prices — is worth paying attention to.
Words worth knowing
Open-weight model — An AI whose inner workings are made publicly available, so others can run it on their own machines instead of renting access from the company that built it.
Context window — How much information an AI can hold in mind at once. A million tokens is roughly 750,000 words — think an entire year of business emails read in a single sitting.
Self-hosting — Running software on your own server or computer instead of using someone else's cloud. More control, usually more effort.
API pricing — When you use an AI through a service, you often pay per chunk of text processed. Lower prices here mean longer conversations cost less.
Something worth sitting with: if powerful AI tools keep getting cheaper and more self-hostable, the competitive edge won't come from access — it'll come from knowing what to do with them.