Open source
Nemotron Transcribe on Nativ: speech understanding on Apple silicon
NVIDIA’s Nemotron Transcribe runs on Nativ through MLX Audio: transcription, speakers, translation and spoken questions, at 33.8× real time on an M3 Ultra.
MiniCPM-OCR on MLX-VLM: document parsing on Apple silicon
OpenBMB’s MiniCPM-OCR 1.0 and Jev now run on MLX-VLM, with 16 public checkpoints. An invoice parses in 0.381 s on an M5 Max.
Docling × MLX-VLM: how IBM Research runs document AI 9.2× faster on a Mac
IBM Research’s Docling runs its document models on Apple Silicon through MLX-VLM. A four-page batch drops from 22.45 to 2.45 seconds on an M3 Max.