oMLX is an open-source inference engine for Apple's MLX local AI ecosystem, released under Apache 2.0 and led by its creator and maintainer Jun Kim. Recent coverage includes Kim joining Hugging Face to work full-time on MLX while continuing to lead oMLX, a user running GLM 5.3 Flash on oMLX 0.7.0 on an M5 Ultra 256 at 68.8 tok/s decode and 1,878 tok/s prefill, and the launch of open-source engine Inco Splash, which claims 144 tok/s for Qwen3.8-27B on an M5 Max, twice oMLX's decode speed.
M5 Ultra 256 runs GLM 5.3 Flash at 68.8 tok/sA user reports running GLM 5.3 Flash with oQ4e+MTP quantization on oMLX 0.7.0 on an M5 Ultra 256, achieving 68.8 tok/s decode and 1,878 tok/s prefill, with more optimization to come. The result responds to earlier benchmarks that were lower than expected.
oMLX creator Jun Kim joins Hugging FaceoMLX creator and maintainer Jun Kim has joined Hugging Face to work full-time on Apple's MLX local AI ecosystem. oMLX remains Apache 2.0 with Kim still leading it, and a focus is streamlining the path from transformers model definitions to MLX reference implementations.
Open-source engine Splash hits 144 tok/s on M5 MaxOpen-source inference engine Inco Splash launched, claiming 144 tok/s for Qwen3.8-27B on an M5 Max MacBook Pro, up to 3x Ollama and 2x oMLX decode speed. It requires M3 or newer, macOS 26.4+, and 36GB RAM, works with Claude Code and Codex, and is available in LM Studio.
The A.I. Daily