The original FP16 (16-bit float) model is ~1.5 GB. After GGML quantization, ggml-medium.bin shrinks to ~500–700 MB . This is the "medium" sweet spot—small enough to run on a Raspberry Pi 4 or an old laptop, but accurate enough for professional-grade transcription.
At its core, ggml-medium.bin is a serialized weight file for the automatic speech recognition (ASR) model, specifically formatted for use with the GGML library. To break that down: ggml-medium.bin
: A multi-lingual model capable of both transcription and translation into English. 2. Performance and Use Cases The original FP16 (16-bit float) model is ~1
It is the embodiment of local-first AI: 769 million integers, carefully packed into half a gigabyte, ready to turn your microphone’s raw audio into clean, timestamped text. For developers and hobbyists, it remains one of the most cost-effective “medium” models ever created. At its core, ggml-medium