The Information Machine

DAY 0 ALERT: @MiniMax_AI M3 is now available on HuggingFace & has been added to InferenceX. The M3 architecture has …

SemiAnalysis Twitter · SemiAnalysis (@SemiAnalysis_) · 2026-06-13

MiniMax AI's M3 model launches on HuggingFace and InferenceX on day 0, featuring roughly 428B total parameters and 23B activated parameters with already-optimized NVIDIA inference performance from Inferact engineers.

Open original ↗

Appears in

Extraction

Topics: minimax-m3model-releasemixture-of-expertsinference-optimization

Claims

  • MiniMax M3 has approximately 428B total parameters with approximately 23B activated parameters, consistent with a mixture-of-experts architecture.
  • M3 is available on HuggingFace and has been added to InferenceX on day 0 of release.
  • Inferact engineers have already achieved well-optimized inference performance on NVIDIA hardware for M3.

Key quotes

DAY 0 ALERT: @MiniMax_AI M3 is now available on HuggingFace & has been added to InferenceX. The M3 architecture has ~428B parameters and ~23B activated parameters.