# MLX 75 Hacker News commenters recommended MLX and 24 criticized it, November 2025 to November 2026. It was recommended in 12 of 13 months. It meets the Hacker Approved badge rule. **Verdict:** Apple's free machine learning framework, recommended for running models fastest on Apple Silicon Macs. Critics point to bugs and to its lack of Neural Engine or matrix hardware use. ## Why people recommend it - Faster than llama.cpp and GGUF (8 comments): "MLX versions specifically optimized for Apple Silicon will also give you noticeably better tok/s" ([source](https://news.ycombinator.com/item?id=47203255)) "supports inference via mlx, which has better performance on apple silicon than llama.cpp" ([source](https://news.ycombinator.com/item?id=47790220)) "make sure to use the MLX over the GGUF version! You'll get a bit more speed out of it." ([source](https://news.ycombinator.com/item?id=47492996)) - Fast updates for new models (3 comments): "the MLX folks and the community usually have MLX versions of new model releases up within a day or two" ([source](https://news.ycombinator.com/item?id=48856174)) "Converting a model to mlx is literally a git clone and mlx_lm.convert. 5 minute operation after clone." ([source](https://news.ycombinator.com/item?id=48857919)) "has some of the fastest updates upon the release of new models" ([source](https://news.ycombinator.com/item?id=48984710)) - Built for Apple hardware (2 comments): "Apple is really seeing huge success with MLX, with an adoptable software stack that the PC world is super struggling to deliver." ([source](https://news.ycombinator.com/item?id=47236490)) "MLX is faster because it has better integration with Apple hardware." ([source](https://news.ycombinator.com/item?id=47585472)) - Large memory and clustering support (2 comments): "You can take advantage of larger amounts of memory, higher memory bandwidth, and clustering multiple systems." ([source](https://news.ycombinator.com/item?id=49514710)) "Or a single mlx cluster if one can find second hand machines somewhere" ([source](https://news.ycombinator.com/item?id=48169721)) ## What people warn about - Bugs and crashes (3 comments): "I do find that it reliably crashes on the first try, but then it picks up from where it finishes and completes the conversion reliably." ([source](https://news.ycombinator.com/item?id=48994181)) "bugs keep getting piled up like they don't care if MLX is actually used in practice" ([source](https://news.ycombinator.com/item?id=47301792)) "the MLX library uses loads of undocumented properties in metal which isn't cool" ([source](https://news.ycombinator.com/item?id=46855325)) - Lower quality at low bits (2 comments): "MLX uses pretty simple quantization methods so at lower bits without QAT, it is just not as good quality as llama.cpp ones." ([source](https://news.ycombinator.com/item?id=49789087)) "I found MLX variants to perform consistently worse (in terms of expected output, not speed) than GGUF" ([source](https://news.ycombinator.com/item?id=47658334)) - Slower in some tests (3 comments): "MLX in the forms I have tried (LM Studio, oMLX) with the models I have tried (Gemma and Qwen) have both been apparently slower, yes." ([source](https://news.ycombinator.com/item?id=48518405)) "I've found them to consistently reduce performance, contrary to what I've read online" ([source](https://news.ycombinator.com/item?id=48211003)) "MLX models have a higher time to first token (sometimes by an order of magnitude)" ([source](https://news.ycombinator.com/item?id=47804628)) - No Neural Engine or matrix hardware (2 comments): "MLX on the software layer will not magically place hardware matrix multiplication units in your GPU cores." ([source](https://news.ycombinator.com/item?id=49481300)) "MLX was and will not be able to access the ANE, so it's a total non-starter for anything user-facing" ([source](https://news.ycombinator.com/item?id=48461432)) ## Alternatives hackers recommend - [PyTorch](https://hackerapproved.com/p/pytorch.md): 131 recommend - [JAX](https://hackerapproved.com/p/jax.md): 49 recommend - [OpenCV](https://hackerapproved.com/p/opencv.md): 46 recommend - [tinygrad](https://hackerapproved.com/p/tinygrad.md): 40 recommend - [TensorFlow](https://hackerapproved.com/p/tensorflow.md): 20 recommend ## Cite as Hacker Approved, "MLX", https://hackerapproved.com/p/mlx/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).