An M1 benchmark under Omarchy puts oMLX ahead of Apple's MLX
Joshua Warren, of the Omarchy M team, said oMLX reaches 39.7 tokens a second on an M1 MacBook Pro running Omarchy, against 27.3 on Apple's MLX, in his benchmark.
The oMLX run used Omarchy's Vulkan patches, Warren said. Its first token arrived in 1.3 seconds, against 7.4 seconds.
The test used one cold prompt of 477 tokens, generated 128 tokens, and ran Qwen3.5 at 4 bit.
Warren built Coreglass for profiling inference. Its code is in the Coreglass repository.
Coreglass turns each run into 1600×900 frames that can be shared. It starts from Super+Space, and pressing t switches it to the current Omarchy theme.
Warren said he has pull requests open on the Omarchy M kernel to add instrumentation for profiling Apple Silicon inference. For now, the tool monitors interrupt activity for the GPU, he said.
Warren said Omarchy M has been certifying M1 and M2 first, which is why the documentation focuses there, and that work on newer chips includes the A18 in the MacBook Neo.
He said he expected a release for opt-in testing on the M3 and M4 the next day, after power-management work.
Correction
The headline said the benchmark put Omarchy ahead. Both runs were on Omarchy, and the result compares oMLX with Apple's MLX. The figures and plans are now attributed to Warren, and the repository is listed as linked from his post.