{"slug": "coreml", "name": "CoreML", "category": "dev-libs", "category_name": "Libraries and frameworks", "description": "", "recommenders": 3, "commenters": 13, "warners": 4, "score": 1.027, "rising": 0.35, "first": "2025-12-14", "latest": "2026-10-14", "why": "I use CoreML for models designed for other platforms. I port the models to it but it works for that without much trouble.", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 4, "neutral": 7, "compare": 0, "warn": 3}, "spark": [0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 0, 1], "listed": true, "loved": [], "caveat": "", "short_quote": "I use CoreML for models designed for other platforms. I port the models to it but it works for that without much trouble.", "kind": "", "icon": true, "domain": null, "github_repo": null, "official_url": "https://developer.apple.com/machine-learning/core-ml/", "pricing": "", "aliases": ["CoreML", "Core ML"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 801343296, "item_id": 50083956, "url": "https://news.ycombinator.com/item?id=50083956", "by": "whstl", "date": "2026-10-14", "version": null, "stance": "recommend", "quote": "It is very suitable and very straightforward for a crazy large amount of models!", "excerpt": "It is very suitable and very straightforward for a crazy large amount of models! I haven\u2019t even mentioned the ones we use at work! There\u2019s a lot beyond LLMs. I suggest trying!", "context": "", "use_case": "running ML models", "switched_from": "", "story": "We turned EmbeddingGemma 2 into a speech-to-text model"}, {"id": 775316176, "item_id": 48457261, "url": "https://news.ycombinator.com/item?id=48457261", "by": "wahnfrieden", "date": "2026-06-08", "version": null, "stance": "recommend", "quote": "I use CoreML for models designed for other platforms. I port the models to it but it works for that without much trouble.", "excerpt": "I use CoreML for models designed for other platforms. I port the models to it but it works for that without much trouble. MLX is not for end user deployment.", "context": "", "use_case": "running ported models on Apple", "switched_from": "", "story": "Apple Core AI Framework"}, {"id": 757248034, "item_id": 47328002, "url": "https://news.ycombinator.com/item?id=47328002", "by": "pzo", "date": "2026-03-10", "version": null, "stance": "recommend", "quote": "I would most likely bet on CoreML since it have support for ANE", "excerpt": "FWIW this RCLI is only MIT license but their engine MetalRT is commercial. Not sure the license of their models I guess also not MIT. So IMHO this repo is misleading. Not sure why they decided to reinvent the wheel and write yet another ML engine (MetalRT) which is proprietary. I would most likely bet on CoreML since it have support for ANE (apple NPU) or MLX. Other popular repos for such tasks I would recommend: htt", "context": "", "use_case": "on-device ML", "switched_from": "", "story": "Show HN: RunAnwhere \u2013 Faster AI Inference on Apple Silicon"}, {"id": 754867248, "item_id": 47179203, "url": "https://news.ycombinator.com/item?id=47179203", "by": "pzo", "date": "2026-02-27", "version": null, "stance": "recommend", "quote": "better use inference on ANE (Apple Neural Engine) via CoreML rather than Metal", "excerpt": "You probably still better use inference on ANE (Apple Neural Engine) via CoreML rather than Metal - speed will be either similar or even faster on non-pro macbooks or iphones and power consumption significantly better. Metal or even MLX format doesn't have to be the fastest and the only way to access ANE is via CoreML. Can use this library: https://github.com/FluidInference/FluidAudio", "context": "", "use_case": "", "switched_from": "", "story": "Parakeet.cpp \u2013 Parakeet ASR inference in pure C++ with Metal GPU acceleration"}]}], "caveats": [{"label": "", "summary": "", "quotes": [{"id": 756127042, "item_id": 47257940, "url": "https://news.ycombinator.com/item?id=47257940", "by": "mechramc", "date": "2026-03-04", "version": null, "stance": "warn", "quote": "CoreML treats it as a black-box scheduler, stripping away any direct control or ability to train", "excerpt": "Hi HN, It is hard to communicate how frustratingly opaque Apple's hardware stack can be. Everyone targets the Mac's GPU for local models, but there is a dedicated accelerator\u2014the Apple Neural Engine (ANE)\u2014sitting completely dark for LLM workloads. CoreML treats it as a black-box scheduler, stripping away any direct control or ability to train. There are a few real caveats here, but imo the fundamental constraint to u", "context": "", "use_case": "Apple ML framework", "switched_from": "", "story": "Show HN: Orion \u2013 Native Training LLMs on the Apple Neural Engine Without CoreML"}, {"id": 755906320, "item_id": 47244145, "url": "https://news.ycombinator.com/item?id=47244145", "by": "dev_l1x_be", "date": "2026-03-03", "version": null, "stance": "warn", "quote": "ANE and CoreML is kind of meh for most use cases", "excerpt": "MPS is okish, is this is what you mean. ANE and CoreML is kind of meh for most use cases, great for some very specific tasks. https://github.com/hollance/neural-engine/blob/master/docs/ane-vs-gpu.md", "context": "", "use_case": "", "switched_from": "", "story": "Apple introduces the new MacBook Air with M5"}, {"id": 741638928, "item_id": 46352433, "url": "https://news.ycombinator.com/item?id=46352433", "by": "Two_hands", "date": "2025-12-22", "version": null, "stance": "warn", "quote": "Also generally I think CoreML isn't the best.", "excerpt": "Thank you for reading. Also generally I think CoreML isn't the best. The best solution for ORT would probably be to introduce a pure MPS provider (https://github.com/microsoft/onnxruntime/issues/21271), but given they've already bought into CoreML the effort may not be worth the reward for the core team. Which fair enough as it's a pretty mammoth task", "context": "", "use_case": "", "switched_from": "", "story": "ONNX Runtime and CoreML May Silently Convert Your Model to FP16"}, {"id": 741630403, "item_id": 46351900, "url": "https://news.ycombinator.com/item?id=46351900", "by": "smcleod", "date": "2025-12-21", "version": null, "stance": "warn", "quote": "has had a few issues running models with CoreML", "excerpt": "This was an interesting read, thanks for sharing. I've recently been building something that uses Parakeet v2/v3 models, I'm using the parakeet-rs package (https://github.com/altunenes/parakeet-rs) which has had a few issues running models with CoreML (unrelated to the linked post), e.g. https://github.com/microsoft/onnxruntime/issues/26355", "context": "", "use_case": "", "switched_from": "", "story": "ONNX Runtime and CoreML May Silently Convert Your Model to FP16"}]}], "who": [], "tags": [], "use_cases": ["running ml models", "production mobile ml", "on-device model inference", "running ported models on apple", "on-device ml", "apple ml framework"], "compared": [{"slug": "mlx", "name": "MLX", "co": 5, "switched_from": 0, "switched_to": 0}, {"slug": "onnx-runtime", "name": "ONNX Runtime", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "fluidaudio", "name": "FluidAudio", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=46350075", "title": "ONNX Runtime and CoreML May Silently Convert Your Model to FP16", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=50069971", "title": "We turned EmbeddingGemma 2 into a speech-to-text model", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49469786", "title": "Benchmarking Pocket-Scale Inference", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49373456", "title": "Show HN: I trained a 125M model to autocomplete piano on-device", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48449665", "title": "Apple Core AI Framework", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47774971", "title": "Google Gemma 4 Runs Natively on iPhone with Full Offline AI Inference", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {"neutral": 1, "warn": 1}, "2026-01": {}, "2026-02": {"recommend": 1}, "2026-03": {"neutral": 2, "warn": 2, "recommend": 1}, "2026-04": {"neutral": 2}, "2026-05": {}, "2026-06": {"recommend": 1}, "2026-07": {}, "2026-08": {"neutral": 2}, "2026-09": {}, "2026-10": {"recommend": 1}}, "versions": [], "version_unnamed": 0, "family_ref": null, "runs_on": null, "lines": [], "quotes": [{"id": 801343296, "item_id": 50083956, "url": "https://news.ycombinator.com/item?id=50083956", "by": "whstl", "date": "2026-10-14", "version": null, "stance": "recommend", "quote": "It is very suitable and very straightforward for a crazy large amount of models!", "excerpt": "It is very suitable and very straightforward for a crazy large amount of models! I haven\u2019t even mentioned the ones we use at work! There\u2019s a lot beyond LLMs. I suggest trying!", "context": "", "use_case": "running ML models", "switched_from": "", "story": "We turned EmbeddingGemma 2 into a speech-to-text model"}, {"id": 791934240, "item_id": 49495890, "url": "https://news.ycombinator.com/item?id=49495890", "by": "HawtAds", "date": "2026-08-30", "version": null, "stance": "neutral", "quote": "traditional transfer learning like Apple's Core.ML", "excerpt": "The benchmark is nice but it's very much biased towards flagships i.e. not very useful in practice if you are trying to ship production mobile apps. Apple historically is extremely stingy when it comes to RAM and they never bothered giving iPads and iPhones more ram until fairly recently (most likely because of ML demands). Your covid era 10th Gen iPads only have 4GB of RAM for the base models. The Android ecosystem ", "context": "", "use_case": "production mobile ML", "switched_from": "", "story": "Benchmarking Pocket-Scale Inference"}, {"id": 790000352, "item_id": 49375022, "url": "https://news.ycombinator.com/item?id=49375022", "by": "simedw", "date": "2026-08-20", "version": null, "stance": "neutral", "quote": "I\u2019m using Core ML", "excerpt": "The biggest speed improvement came from changing the note representation when I switched to compound note events: roughly 5\u00d7 fewer autoregressive passes per note. For the current model I\u2019m using Core ML, which optimizes the kernels the first time you run it. I haven\u2019t actually spent that much time tuning performance beyond that.", "context": "", "use_case": "on-device model inference", "switched_from": "", "story": "Show HN: I trained a 125M model to autocomplete piano on-device"}, {"id": 775316176, "item_id": 48457261, "url": "https://news.ycombinator.com/item?id=48457261", "by": "wahnfrieden", "date": "2026-06-08", "version": null, "stance": "recommend", "quote": "I use CoreML for models designed for other platforms. I port the models to it but it works for that without much trouble.", "excerpt": "I use CoreML for models designed for other platforms. I port the models to it but it works for that without much trouble. MLX is not for end user deployment.", "context": "", "use_case": "running ported models on Apple", "switched_from": "", "story": "Apple Core AI Framework"}, {"id": 764513312, "item_id": 47782082, "url": "https://news.ycombinator.com/item?id=47782082", "by": "Yukonv", "date": "2026-04-15", "version": null, "stance": "neutral", "quote": "requires using a specific pipeline specifically for CoreML", "excerpt": "It is possible but requires a very specific model design to utilize. As this reverse engineering effort has shown [0] \"The ANE is not a GPU. It\u2019s not a CPU. It\u2019s a graph execution engine.\" To build one requires using a specific pipeline specifically for CoreML [1]. [0] https://maderix.substack.com/p/inside-the-m4-apple-neural-engine [1] https://developer.apple.com/documentation/coreml", "context": "", "use_case": "", "switched_from": "", "story": "Google Gemma 4 Runs Natively on iPhone with Full Offline AI Inference"}, {"id": 764485970, "item_id": 47780373, "url": "https://news.ycombinator.com/item?id=47780373", "by": "mowmiatlas", "date": "2026-04-15", "version": null, "stance": "neutral", "quote": "If anyone on CoreML has a principled explanation", "excerpt": "Extra context since the post got long. A few things that ate more time than I expected: Streaming was the worst one. Kokoro doesn't expose a streaming interface as far as I could find, you hand it a chunk of text, it gives you back the full audio for that chunk. For a reading app you can't wait for a whole paragraph before playback starts, so the whole streaming layer had to be built on top. I didn't want to process ", "context": "", "use_case": "", "switched_from": "", "story": "Show HN: I built on-device TTS app because I run out of audiobooks on a flight"}, {"id": 757248034, "item_id": 47328002, "url": "https://news.ycombinator.com/item?id=47328002", "by": "pzo", "date": "2026-03-10", "version": null, "stance": "recommend", "quote": "I would most likely bet on CoreML since it have support for ANE", "excerpt": "FWIW this RCLI is only MIT license but their engine MetalRT is commercial. Not sure the license of their models I guess also not MIT. So IMHO this repo is misleading. Not sure why they decided to reinvent the wheel and write yet another ML engine (MetalRT) which is proprietary. I would most likely bet on CoreML since it have support for ANE (apple NPU) or MLX. Other popular repos for such tasks I would recommend: htt", "context": "", "use_case": "on-device ML", "switched_from": "", "story": "Show HN: RunAnwhere \u2013 Faster AI Inference on Apple Silicon"}, {"id": 757034898, "item_id": 47314681, "url": "https://news.ycombinator.com/item?id=47314681", "by": "busymichael", "date": "2026-03-09", "version": null, "stance": "neutral", "quote": "Rebuilt as an iOS/macOS app with CoreML", "excerpt": "Building a voice recorder/transcription/summarization app where everything runs on-device. No cloud, no uploads, no servers processing your audio. Started with a web version using WebAssembly Whisper to validate the idea. Worked well enough to prove demand, but browser sandbox limits (no background recording, inconsistent model performance) pushed me to go native. Rebuilt as an iOS/macOS app with CoreML. Apple Silico", "context": "", "use_case": "", "switched_from": "", "story": "Ask HN: What Are You Working On? (March 2026)"}, {"id": 756127042, "item_id": 47257940, "url": "https://news.ycombinator.com/item?id=47257940", "by": "mechramc", "date": "2026-03-04", "version": null, "stance": "warn", "quote": "CoreML treats it as a black-box scheduler, stripping away any direct control or ability to train", "excerpt": "Hi HN, It is hard to communicate how frustratingly opaque Apple's hardware stack can be. Everyone targets the Mac's GPU for local models, but there is a dedicated accelerator\u2014the Apple Neural Engine (ANE)\u2014sitting completely dark for LLM workloads. CoreML treats it as a black-box scheduler, stripping away any direct control or ability to train. There are a few real caveats here, but imo the fundamental constraint to u", "context": "", "use_case": "Apple ML framework", "switched_from": "", "story": "Show HN: Orion \u2013 Native Training LLMs on the Apple Neural Engine Without CoreML"}, {"id": 755906320, "item_id": 47244145, "url": "https://news.ycombinator.com/item?id=47244145", "by": "dev_l1x_be", "date": "2026-03-03", "version": null, "stance": "warn", "quote": "ANE and CoreML is kind of meh for most use cases", "excerpt": "MPS is okish, is this is what you mean. ANE and CoreML is kind of meh for most use cases, great for some very specific tasks. https://github.com/hollance/neural-engine/blob/master/docs/ane-vs-gpu.md", "context": "", "use_case": "", "switched_from": "", "story": "Apple introduces the new MacBook Air with M5"}, {"id": 755764496, "item_id": 47235281, "url": "https://news.ycombinator.com/item?id=47235281", "by": "bri3d", "date": "2026-03-03", "version": null, "stance": "neutral", "quote": "the ANE toolchain from CoreML down through Espresso", "excerpt": "I'm very familiar with CFG flattening and other obfuscation techniques, thanks. That's interesting; I suppose I must not have touched the parts of the platform that use them, and I've touched a fair amount of the platform. Again, I _have_ seen plenty of obfuscation techniques in DRM/FairPlay, but otherwise I have not, and again, I am entirely sure the ANE toolchain from CoreML down through Espresso and into AppleNeur", "context": "", "use_case": "on-device ML models", "switched_from": "", "story": "Inside the M4 Apple Neural Engine, Part 1: Reverse Engineering"}, {"id": 754867248, "item_id": 47179203, "url": "https://news.ycombinator.com/item?id=47179203", "by": "pzo", "date": "2026-02-27", "version": null, "stance": "recommend", "quote": "better use inference on ANE (Apple Neural Engine) via CoreML rather than Metal", "excerpt": "You probably still better use inference on ANE (Apple Neural Engine) via CoreML rather than Metal - speed will be either similar or even faster on non-pro macbooks or iphones and power consumption significantly better. Metal or even MLX format doesn't have to be the fastest and the only way to access ANE is via CoreML. Can use this library: https://github.com/FluidInference/FluidAudio", "context": "", "use_case": "", "switched_from": "", "story": "Parakeet.cpp \u2013 Parakeet ASR inference in pure C++ with Metal GPU acceleration"}, {"id": 741638928, "item_id": 46352433, "url": "https://news.ycombinator.com/item?id=46352433", "by": "Two_hands", "date": "2025-12-22", "version": null, "stance": "warn", "quote": "Also generally I think CoreML isn't the best.", "excerpt": "Thank you for reading. Also generally I think CoreML isn't the best. The best solution for ORT would probably be to introduce a pure MPS provider (https://github.com/microsoft/onnxruntime/issues/21271), but given they've already bought into CoreML the effort may not be worth the reward for the core team. Which fair enough as it's a pretty mammoth task", "context": "", "use_case": "", "switched_from": "", "story": "ONNX Runtime and CoreML May Silently Convert Your Model to FP16"}, {"id": 741630403, "item_id": 46351900, "url": "https://news.ycombinator.com/item?id=46351900", "by": "smcleod", "date": "2025-12-21", "version": null, "stance": "warn", "quote": "has had a few issues running models with CoreML", "excerpt": "This was an interesting read, thanks for sharing. I've recently been building something that uses Parakeet v2/v3 models, I'm using the parakeet-rs package (https://github.com/altunenes/parakeet-rs) which has had a few issues running models with CoreML (unrelated to the linked post), e.g. https://github.com/microsoft/onnxruntime/issues/26355", "context": "", "use_case": "", "switched_from": "", "story": "ONNX Runtime and CoreML May Silently Convert Your Model to FP16"}, {"id": 740290625, "item_id": 46268164, "url": "https://news.ycombinator.com/item?id=46268164", "by": "Two_hands", "date": "2025-12-14", "version": null, "stance": "neutral", "quote": "when using the CoreML execution provider", "excerpt": "Working on a blog post about default behaviour in ONNX Runtime when using the CoreML execution provider. Basically the default args lead to your model being ran in FP16 not FP32. You can find more details at my site soon: https://ym2132.github.io", "context": "", "use_case": "", "switched_from": "", "story": "Ask HN: What Are You Working On? (December 2025)"}]}