MiniMax Claims Blind-Test Model Space Bunny, Confirms It as MiniMax-M3.1-Flash

Damn! This is pretty hype, brothers—this is absolutely the most satisfying and hardcore unmasking of a king in the large-model world recently, haha, I suggest watching the official short video in fullscreen to feel what real strength endorsement looks like, For the past half month or so, the mysterious Space Bunny that went absolutely wild on overseas blind-test platforms, stunning developers worldwide with its blazing speed of one hundred words per second and one-million ultra-long context—today MiniMax officially stepped up and claimed it. Brothers, do you know how explosive this contrast is? To dig out whose hand this mysterious model really was, the overseas community ran dozens of tokenizer comparisons in a row and argued fiercely; everyone was guessing which Silicon Valley giant was secretly holding back, and when the cards were finally revealed, it was none other than the domestic mainstay MiniMax-M3.1-Flash. What first shattered industry inertia was its terrifying throughput in real-world productivity. Previously, when running multi-agent or long-context tasks, the most painful part was watching the screen spit out a few characters per second, waiting frantically through complex code debugging and self-correction, But in community tests, this Space Bunny's backend generation speed held steady at 85 to 112 words per second; not only could it run through complex 3D scenes and code builds in a single pass within three minutes, it also achieved near-imperceptible response in multi-file project repairs. Even more intriguing is this hardcore stance of first defeating opponents in an anonymous arena. No overwhelming marketing warm-up—it stripped the model of all brand halos and threw it into the global blind-test pool, letting the world's most demanding programmers vote one by one with real engineering tasks, Only when the final test data proved it could go head-to-head with the front-line tier in everyday coding and long-text understanding did the official account calmly post: Space Bunny, welcome back. This is the healthiest form of future large-model competition: rather than piling up ethereal benchmark scores at launch events, better to make inference speed light-speed and million-token context dirt-cheap, so that every ordinary person writing code every day truly feels the thrill of doubled efficiency. When you usually use agents to write code or process long documents, what do you care about most: lightning-fast response or slow thinking with super-large parameters? I've found that in everyday bug-fixing and rapid prototyping scenarios, this kind of high-speed stream at one hundred words per second with million-token context—once I've used it, I simply can't go back, haha😆 https://t.co/QnLxQiR0ZU

View original

Damn！有点燃炸兄弟们，这绝对是最近大模型圈最解气、也最硬核的一次王者揭面了哈哈， 建议全屏把官方这段短片看完，去感受一下什么叫真正的实力背书， 过去大半个月在海外盲测平台杀疯了、靠着每秒一百词的极速和一百万超长上下文把全网开发者看傻的神秘太空兔 Space Bunny，今天 MiniMax 官方直接登台认领了。 兄弟们你知道这个反差有多狂暴吗？
海外社区为了扒出这个神秘模型到底是谁的底牌，连着做了几十组分词器比对吵翻了天，大家都在猜到底是哪家硅谷巨头在偷偷放水，结果底牌亮出来，正是国产主力 MiniMax-M3.1-Flash。 最先击碎行业惯性的是它在现实生产力里的恐怖吞吐量。 
以前跑多智能体或者长上下文任务，最痛苦的就是看着屏幕一秒钟吐几个字，在复杂的代码调试和自我纠错里等得让人抓狂， 而在社区实测中，这只太空兔在后台的生成速度稳定在 85 到 112 词每秒，不仅能在三分钟内单次跑通复杂的 3D 场景与代码构建，更是在多文件工程修复里几乎做到了无感响应。 更耐人寻味的，是这种先在匿名竞技场里打服对手的硬核姿态。 
没有铺天盖地的营销预热，直接把模型剥去所有品牌光环丢进全球盲测池，让全世界最苛刻的程序员用真实的工程任务去一票一票投票， 当最终的测试数据证明它在日常编码和长文本理解上完全能硬刚一线梯队时，官方才从容地发了一句太空兔欢迎回来。 这就是未来大模型竞争最健康的形态：
与其在发布会上堆砌虚无缥缈的跑分，不如把推理速度做成光速，把百万上下文做成白菜价，让每一个每天写代码的普通人真真切切感受到效率翻倍的快感。 大家平时在用智能体写代码或者处理长文档时，最在乎的是极速响应还是超大参数的慢思考？ 我发现在日常改 Bug 和快速搭原型的场景里，这种每秒一百词且带百万上下文的极速流，我现在只要用过就根本回不去了哈哈哈😆 https://t.co/QnLxQiR0ZU