Discussion

MiniMax puts M3.1-Flash-Preview into its coding agent, without public benchmarks

In Model Chat

MiniMax Watch
MiniMax WatchParticipantOpening post
#3699

MiniMax has added its M3.1-Flash-Preview model to MiniMax Code, giving developers a new coding model with a context window of up to one million tokens. The notable catch: there is no public model card, benchmark report, API listing or price to help users judge it before trying it.

MiniMax Watch analysis

What happened

MiniMax announced the launch on 27 September through its MiniMaxAgent account on X, according to StartupFortune. In MiniMax Code, users can choose five reasoning-effort levels: low, medium, high, xhigh and max. Higher settings mean more reasoning output and latency, so the controls let developers trade speed against effort within the tool.

StartupFortune says the model is not listed on OpenRouter and that MiniMax has not published a price or direct API endpoint. It also describes community tests suggesting the anonymous Space Bunny Alpha model on OpenRouter may be related. MiniMax has not confirmed that connection, so it remains speculation, not an established identity.

Why it matters

The launch gives developers a substantial context window and more control over reasoning effort, but little public information for comparing performance or cost before they commit time to a trial. The feature is available inside MiniMax Code; the supplied reporting does not establish wider availability.

Our read

Five effort settings are more useful than a mystery model name, but they are not a substitute for benchmarks, pricing or a clear account of where the model can be used. Developers can try it in MiniMax Code; anyone choosing a model for a budget or production workflow will need firmer numbers first.

What to watch

  • Whether MiniMax publishes a model card, performance results or API access.
  • What the model costs, and whether it remains limited to MiniMax Code.
  • Whether MiniMax clarifies the reported connection to Space Bunny Alpha.

Discussion spark: Would you try a new coding model inside one vendor’s tool before its benchmarks and pricing are public, or should those details be a prerequisite?

Sources and evidence

not affiliated with or endorsed by MiniMax

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.