Space Bunny（M3.1 Flash 匿名版）登顶 OpenRouter 后揭晓
匿名模型 Space Bunny 连续十天在 OpenRouter 排名第一，处理 56.7 万亿 token；MiniMax 确认它是 M3.1 Flash 早期版，并公布匿名期反馈带来的改进：首 token 时间降 38%、吞吐 70→100 tok/s、错误率减半，以及厂商跑分和 72 小时免费。
The most mysterious model on OpenRouter just took off its mask. For weeks, an anonymous model called Space Bunny sat at #1 on OpenRouter's leaderboard. Ten straight days. 56.7 trillion tokens processed, more than any stealth model they have ever hosted. Nobody knew whose it was. Today MiniMax said it was theirs. Space Bunny was an early version of M3.1 Flash, which launched this morning. 200 trillion tokens in stealth, by MiniMax's count. The stealth run was not just marketing. OpenRouter says user feedback during those anonymous weeks helped MiniMax fix real bugs. Time to first token dropped 38%, 1.95 seconds to 1.22. Throughput went from 70 to 100 tokens per second. Error rate halved, 1.4% to 0.7%. The model itself: 3x faster inference than its predecessor at 70% lower cost. $0.15 per million input tokens. Vendor-reported benchmarks: DeepSWE v1.1 73.7%, OSWorld V2 61.7, CyberGym 82%, BankerToolBench 76.2, BabyVision 91%. First among Chinese models on OSWorld V2 and Agents' Last Exam, per MiniMax. It is free on OpenRouter for 72 hours. Weights are promised next. A stealth launch that actually used the stealth period for something. I like it. https://www.unite.ai/minimax-launches-m3-1-flash-revealing-space-bunny-stealth-model/