MiniMax M3.1-Flash-Preview lands in MiniMax Code: 428B MoE (23B active), 1M context, native vision encoder — free to try in the desktop app
Developer @exploraX_ posted a hands-on video: MiniMax's latest text model M3.1-Flash-Preview is now live in the MiniMax Code desktop app (macOS/Windows), free to use after signing in with a Google account, plus daily check-in credits; subscribers get unlimited use Oct 1-7. Official specs: MoE with 428B total / ~23B active parameters, sparse attention, 1M-token context, ~150 tokens/s, and a native visual encoder accepting text/image/video input. Reasoning is always on (cannot be disabled) with five effort levels (Low/Medium/High/Xhigh/Max; Max is the recommended default). Positioned for everyday development — quick bug fixes to full features — plus financial research, multimodal understanding and frontend work; official examples include three-statement+DCF modeling, image-difference analysis, a procedural Three.js building (Qinian Hall) generated from reference photos, and interactive fluid simulation. API model id: MiniMax-M3.1-Flash-Preview. Key change from M3: reasoning always on + configurable reasoning effort. MiniMax also launched M Plan (50% off first month until Oct 14).
