MiniMax Unmasks Space Bunny as M3.1 Flash After 200T Tokens
MiniMax has officially released MiniMax M3.1 Flash, solving one of the community's favorite recent mysteries. The Chinese AI lab confirmed the high-speed system was the identity behind Space Bunny, the viral stealth model that dominated global evaluation boards and API routers over the past month.
The mystery model that crushed OpenRouter
For weeks, builders experimenting on model aggregators noticed a rapid, unusually capable entry disguised as Space Bunny. The mystery model shot directly to the top of developer boards, outperforming established frontier options on latency while maintaining surprisingly sharp reasoning.
Aggregator platform OpenRouter revealed that Space Bunny held its number-one spot for 10 consecutive days before quietly disappearing on October 5. During that test run alone, it processed 56.7 trillion tokens—setting an all-time usage record for any anonymous stealth deployment hosted on the platform. MiniMax noted that total pre-release evaluation traffic topped 200 trillion tokens across global arenas.
MiniMax M3.1 Flash: specs, usage and availability
| Spec | Detail |
|---|---|
| Developer | MiniMax |
| Stealth Alias | Space Bunny |
| Release Date | October 19, 2026 |
| Pre-launch Volume | 200T+ tokens |
| OpenRouter Stealth Volume | 56.7T tokens |
| Launch Promo | Free for 72 hours via OpenRouter |
Targeting speed and cost in agentic workflows
MiniMax built the Flash edition specifically to handle autonomous developer loops.
MiniMax is positioning MiniMax M3.1 Flash as an agent-focused engine designed to deliver fast response times without degrading to basic classification accuracy. While flagship frontier models often drag their feet through multi-step agent loops, lightweight 'flash' models trade away tiny margins of fringe knowledge for raw throughput.
The release fits a broader industry push to bring operational token expenses down, especially as enterprises rethink ballooning inference costs. Other labs have taken hardware shortcuts—similar to how Microsoft's Maia 300 bets on memory bandwidth to curb inference overhead—while MiniMax chose architectural pruning to push execution speeds.
Stealth model drops are the best marketing trick in modern AI, but Space Bunny proved something rare: it actually held up under heavy developer scrutiny. 56 trillion tokens on OpenRouter is not an artificial synthetic benchmark; developers kept calling it because it was genuinely fast and capable. The question now is whether MiniMax can maintain that latency once enterprise agent loops hit production volumes.
Availability and pricing
MiniMax M3.1 Flash is accessible immediately worldwide through MiniMax's platform and third-party API aggregators. To mark the unmasking, OpenRouter is offering the model completely free for the first 72 hours of launch, allowing developers in North America and Europe to test their agent workflows against production traffic without upfront billing.
With the rabbit out of the hat, MiniMax transitions from an intriguing blind test to commercial competition against the established flash tiers from Google and Anthropic. If the production API matches the stealth performance, developers building autonomous agents have a compelling high-speed contender.
Frequently asked questions
What was the Space Bunny AI model?
Space Bunny was a stealth model released anonymously on AI testing leaderboards and OpenRouter. MiniMax confirmed it was an early evaluation build of MiniMax M3.1 Flash.
How can developers access MiniMax M3.1 Flash?
MiniMax M3.1 Flash is available directly via the MiniMax API and OpenRouter, where it launched with a 72-hour free usage window.
Sources
- MiniMax Unveils M3.1 Flash AI Model, Previously Tested as 'Space Bunny'. X gündemi: Space Bunny, 2026. https://x.com/MiniMax_AI/status/2112150867868541221