M3.1 Flash 预告汇总：1M 上下文、5 档思考深度、早期测试褒贬不一
整理 MiniMax 官方预告与已知规格：多模态编程模型、1M token 上下文、5 档思考深度、兼容 OpenAI/Anthropic API；早期测试者称快且便宜，也有人认为预览版不如 DeepSeek V4.1 Flash 少错。
MiniMax just teased MiniMax-M3.1-Flash with a Space Bunny video today. Here's what you need to know. MiniMax's official X account posted "Welcome back. Space Bunny. 🚀 MiniMax-M3.1-Flash is coming." It came with a 25-second video. RyanLee of MiniMax had posted "Dear friends, we are back." an hour earlier. The post passed 500K views in about 12 minutes and now sits at 545K. Space Bunny was a stealth model that appeared free on OpenCode and OpenRouter. Testers guessed it was MiniMax from its tokenizer and benchmark runs. On Sep 26, LuminaBench noted MiniMax's "M3-a" deployment returned the model name "space-bunny". MiniMax has now confirmed it is M3.1-Flash. Users say it became the most used model on OpenCode and OpenRouter while free. What we know about the model. MiniMax's API docs list M3.1-Flash as a multimodal coding model with a 1M-token context window and tunable thinking depth. It takes text and image input, and reports mention video input too. It has OpenAI and Anthropic-compatible APIs. AIHubmix says it is live there for coding, office automation and visual reasoning. Early testers call it fast and very cheap, though MiniMax hasn't published per-token pricing in the material we have. Reactions are mixed. Some early users praise its speed in agent coding loops. One tester found the preview version disappointing, and another found DeepSeek V4.1 Flash made fewer errors than Space Bunny. MiniMax's Zekun also noted the model is in the API docs. On Oct 17 Zekun posted that more open releases are coming. Key numbers: - 1M-token context window - 5 thinking effort levels in the preview - Teaser video: about 25 seconds - Teaser views: 545K, with 3.8K likes - Token Plan subscriptions start at $22 a month MiniMax says M3.1-Flash is coming, with no exact release date given yet.
这条是引用着另一条写的：「MiniMax 官方预告 M3.1 Flash 发布，点名 Space Bunny」