Models & LLMsAI Infrastructure

Space Bunny Alpha Is MiniMax M3.1 Flash: What Changed

OpenRouter revealed that the Space Bunny Alpha stealth model is MiniMax M3.1 Flash. The free launch ended Oct 22, 2026 at 12:00 UTC, and the base slug now costs $0.15 in and $0.60 out per million tokens.

Toolbit AI - Team
4 min read
Space Bunny Alpha Is MiniMax M3.1 Flash: What Changed

OpenRouter's Space Bunny Alpha page states that the stealth model was built by MiniMax and is MiniMax M3.1 Flash. The free stealth period has ended, and the free launch of the named model ended on Oct 22, 2026 at 12:00 UTC. After that, OpenRouter serves the base slug at MiniMax list price: $0.15 per million input tokens and $0.60 per million output tokens.

What happened to Space Bunny Alpha?

The stealth page says the model "was developed and operated by MiniMax, revealed to be MiniMax M3.1 Flash." It adds that OpenRouter only routed requests and was not the developer. The stealth listing went up on Sep 23, 2026, and the named MiniMax M3.1 Flash listing followed on Oct 19, 2026.

What changed: the stealth model has a name, MiniMax M3.1 Flash, the free stealth period is over, and the free launch ended Oct 22, 12:00 UTC. Who's affected: anyone calling stealth/space-bunny-alpha or the :free variant of M3.1 Flash. What to do: switch to minimax/minimax-m3.1-flash now, and budget for the list price.

How does the named model compare with the stealth listing?

The context window stayed at 1M tokens, while the price and the provider name changed. MiniMax's pay-as-you-go pricing page lists the same rates OpenRouter charges after the free launch.

Space Bunny Alpha (stealth)MiniMax M3.1 Flash
Model idstealth/space-bunny-alphaminimax/minimax-m3.1-flash (MiniMax API: MiniMax-M3.1-Flash)
Listed on OpenRouterSep 23, 2026Oct 19, 2026
ProviderAnonymous, now revealed as MiniMaxMiniMax
Context1.0M1.0M
InputsNot restatedMultimodal (OpenRouter listing)
Price per M tokensFree (period ended)Free launch ended Oct 22, 12:00 UTC; now $0.15 input, $0.60 output, $0.015 cached input

On Oct 21 the page we read still showed the :free variant at "Free", and OpenRouter said that variant would be retired after the Oct 22 deadline.

What it means for developers on the stealth slug

Your stealth endpoint is done, so the model id in your client has to change. OpenRouter's instruction is direct: "switch to the base slug to avoid interruption."

On MiniMax's own API, MiniMax's docs say effort "accepts low, medium, high, xhigh, and max," and that omitting it defaults to max for M3.1 Flash. Thinking is on by default. Use output_config.effort on the Anthropic-compatible API or reasoning_effort on the OpenAI-compatible one, where thinking comes back in reasoning_content.

Data you already sent stays under the stealth terms. That listing said prompts and completions "may be retained by the provider but are not used for training."

What it means if you watch cost

M3.1 Flash is cheaper than MiniMax M3, which OpenRouter lists at $0.23 input and $0.96 output per million tokens. Take a sample job of 1M input tokens and 200k output tokens.

Total cost of one sample job, 1M input plus 200k output tokens: $0.162 on M3.1 Flash with 800k cached, $0.27 uncached, $0.422 on MiniMax M3
  • M3.1 Flash, no cache: $0.15 + $0.12 = $0.27.
  • M3.1 Flash, 800k of the input as cached reads: $0.03 + $0.012 + $0.12 = $0.162.
  • MiniMax M3: $0.23 + $0.192 = $0.422, so M3.1 Flash uncached costs about 36% less.

If you plan to pick a model per task rather than switch wholesale, compare it on the MiniMax M3 page and our guide to choosing a model by task.

What to do now

  1. Now: replace stealth/space-bunny-alpha and minimax/minimax-m3.1-flash:free with minimax/minimax-m3.1-flash.
  2. Same day: pin that id in config so a later listing change can't silently reroute you.
  3. Before you scale up: price your usual workload on OpenRouter at $0.15 input and $0.60 output per million, with and without cached reads at $0.015.
  4. After the switch: set effort explicitly if you don't want max, since omitting it defaults to max.

How should you run a short eval at list price?

The free window is over, so any bake-off now runs at list price. Keep it small and throwaway before you move a whole workload onto the model.

  1. Sweep the same task set across low, medium, high, xhigh and max.
  2. Log latency, token burn and tool-call success yourself.
  3. Keep prompts synthetic or public. No secrets, customer data or regulated data.

Still unconfirmed

  • Open weights: OpenRouter now describes M3.1 Flash as an open-weight model and links the weights on Hugging Face; check the license there before self-hosting.
  • Max output tokens: OpenRouter now lists up to 1,048,576 completion tokens; providers may cap lower, so check the one you route to.
  • Benchmarks: the scores on OpenRouter come from Artificial Analysis, a third party, and have changed since launch, so read the current figure there. MiniMax's claim of "significant improvements" over M3 is the vendor's own.

See the MiniMax M3.1 Flash model page for specs as they update.

Share this article

Related articles

Continue exploring similar guides and insights