MiniMax has released M3.1 Flash and confirmed that the previously anonymous Space Bunny model was an early version of the new system. The Chinese AI startup said M3.1 Flash brings what it describes as generational improvements in agentic tasks, programming, and cybersecurity.
The release also ends weeks of speculation among developers about Space Bunny’s origin. MiniMax previewed the announcement on X with the message, “Welcome back. Space Bunny. MiniMax-M3.1-Flash is coming.” The company said Space Bunny processed more than 200 trillion tokens while it operated in stealth.
M3.1 Flash Pricing and Availability
MiniMax lists flat API pricing for M3.1 Flash at:
- $0.15 per million input tokens
- $0.60 per million output tokens
- $0.015 per million cached-read tokens
According to MiniMax, M3.1 Flash runs at one-third of the cost of its M3 model while delivering three times the speed. The company also said online inference speed has nearly tripled and usage costs have fallen by 70%.
M3.1 Flash is available through the MiniMax API, the M Plan subscription, and the MiniMax Code desktop app. OpenRouter is making the model available at no cost for 72 hours before moving to MiniMax API-equivalent pricing. OpenCode is also offering the model free for three days.
M3.1 Flash Input and Context Window
The model supports multimodal input and has a context window of 1 million tokens. MiniMax has not disclosed its parameter count or license.
MiniMax said it plans to release the model’s open weights on October 23.
Space Bunny’s Stealth Run on OpenRouter and OpenCode
Space Bunny Alpha first appeared on OpenRouter on September 23 without an identified developer. By September 24, it had reached the top of OpenCode’s daily token-usage ranking.
OpenCode said Space Bunny held first place on its platform for 12 consecutive days and served 106 trillion tokens. MiniMax separately said that the model processed more than 200 trillion tokens during its stealth period.
During that run, OpenRouter said MiniMax improved several performance measures:
|
Metric
|
Earlier Result
|
Later Result
|
|
Time to first token
|
1.95 seconds
|
1.22 seconds
|
|
Throughput
|
70 tokens per second
|
100 tokens per second
|
|
Error rate
|
1.4%
|
0.7%
|
MiniMax M3.1 Flash Benchmark Results
In MiniMax’s own comparison table, M3.1 Flash recorded a 62.2% score on Automation-Bench. That result was higher than DeepSeek-V4.1-Flash-1017 at 60.5% and Claude Opus 5 at 50.3%.
The model did not lead every reported comparison. On the DeepSWE coding test, M3.1 Flash scored 73.7%, compared with 77.0% for DeepSeek. On the BabyVision visual-reasoning test, M3.1 Flash reached 91.0%, while Claude Opus 5 scored 94.1%.
A Higher-End MiniMax Pro Model Is Planned
MiniMax said a higher-end Pro version is expected later. The company described the upcoming model as one that will push the boundaries of intelligence.
M3.1 Flash, meanwhile, is positioned as part of MiniMax’s effort to balance efficiency and capability.
FAQ
What is Space Bunny?
Space Bunny was an anonymous model used by developers before MiniMax confirmed it was an early version of MiniMax M3.1 Flash. It appeared on OpenRouter as Space Bunny Alpha during a stealth run.
How much does MiniMax M3.1 Flash cost?
The API costs $0.15 per million input tokens and $0.60 per million output tokens. Cached reads cost $0.015 per million tokens.
When will MiniMax release M3.1 Flash open weights?
MiniMax said it will release open weights for M3.1 Flash on October 23. The parameter count and license have not been disclosed.