minimax m3.1

MiniMax M3.1 Flash appears to refer to a multimodal MiniMax AI model for coding, agentic workflows, computer use, and cybersecurity, with text/image/video inputs and text output. Sources consistently describe it as a 1M-context MoE model with tunable reasoning depth, though some third-party listings differ on parameter counts and availability details.[1][2][15]

The public sources establish that MiniMax-M3.1-Flash has about 428B total parameters with roughly 23B active per token, supports up to a 1,000,000-token context window, and uses reasoning levels from low to max.[1][2][19] They also show pricing in MiniMax’s docs and mention an open-weights release, but release timing, exact access routes, and some spec fields vary across providers, so those details should be treated as source-dependent rather than fully settled.[12][16][24]

perplexity/sonar · Prepared 2026-10-29 02:07 UTC · Check the original sources; information may change.