Model ids are now the real model names
We dropped our invented model aliases — what you request is what runs, under the name its maker gave it.

When we launched, the catalog used Hyphen-branded aliases that mapped onto whatever we routed to underneath. It felt tidy — stable names, swappable backends. It also meant you couldn't tell at a glance which model was actually answering, and we've decided that's the wrong trade. If you point an agent at a gateway, you deserve to know exactly what's on the other end.
So the aliases are gone. What you request is what runs: catalog ids are the real provider model names, and every API response reports the model that actually served the request. The catalog will keep changing over time — models added, upgraded, retired — and we'll give notice when it does. The one exception is incidents: if a provider is having a bad day, a request may fall back to a smaller model of the same family, and the response will say so.
At rename time that meant four ids: minimax-m3, minimax-m2.7, minimax-m2.5, and minimax-text-01. Today the catalog is ten MiniMax models — those four plus highspeed and specialty variants — every one listed under its provider name.
If you were calling a hyphen-* alias, it no longer exists: point your config at a real id from the catalog instead. Everything else — the endpoint, your key, your plan — is unchanged.
Honesty wins here for a boring reason: you can't debug what you can't see. Real names make your evals, your cost intuition, and your bug reports all line up with reality.
- minimax-m3Flagship. MiniMax's strongest — long-horizon agentic coding, tool use, and deep reasoning.
- minimax-m2.7Balanced daily driver. ~205K context; fast and strong across coding and agents.
- minimax-m2.5Fastest of the family. Quick and cheap — ideal for high-volume, lighter calls.
- minimax-text-01Long-context text model. Feed it whole codebases or big document sets in one prompt.