Space Bunny (M3.1 Flash) specs and hands-on: strong frontend, shaky complex state management
A rundown of official specs — 1M context, text/image/video input, five effort levels defaulting to max, thinking always on — plus hands-on notes that frontend work nears flagship level while complex interactions and state management remain shaky, with speed and value as clear wins.
That viral Space Bunny Alpha on OpenRouter / OpenCode? It’s MiniMax-M3.1-Flash — about to ship for real. Specs (from the official docs): Context: 1,000,000 tokens Multimodal in: text, image, video Effort: low / medium / high / xhigh / max (default max) Thinking: on by default, can’t turn it off Aimed at agent reasoning, tool use, coding, long-context — frontier coding model APIs: Anthropic-compatible and OpenAI-compatible It’s already in the official API docs. Devs can try it via MiniMax’s M Plan / MiniMax Code. Full pricing and wider rollout still TBD. Another MiniMax coding push after M3 — fast, multimodal, tunable thinking depth. Space Bunny’s unmasked. Looking forward to real benches once it’s fully live. Frontend / visual gen is strong — close to flagship, and sometimes matches it. Day-to-day coding and agent work are usable, but complex interactions and state management still get shaky. Speed and bang-for-buck are the clear wins.
This post quotes another post: “MiniMax teases M3.1 Flash launch and nods to Space Bunny”