Skip to content

On-device models

Models you can run on iPhone, iPad and Mac, from Hugging Face

Core ML and MLX models published on Hugging Face, plus the models Apple publishes there, sorted by downloads in the last 30 days. Core ML models run on iPhone, iPad and Mac through Apple's Core ML framework; MLX models are made for Apple's MLX library, mostly on Apple silicon Macs (MLX Swift also runs on iPhone and iPad). Each name links to the model's page on Hugging Face.

1,009 Core ML models, 1,693 MLX models and 143 from Apple, updated daily from Hugging Face's public API. Last updated 2026-11-02 (UTC).

New this month

The most liked models created in the last 30 days, across Core ML, MLX and Apple's own.

See all new models

Showing 100 of 100.

ModelTaskSizeDownloads (30 days)
prism-ml/Ternary-Bonsai-27B-mlx-2bitMLXfrom Qwen/Qwen3.6-27BText generation8.5 GB985K
prism-ml/Bonsai-27B-mlx-1bitMLXfrom Qwen/Qwen3.6-27BText generation5.2 GB976K
rapid-mlx/Ling-3.0-tiny-MLX-4bitMLXfrom inclusionAI/Ling-3.0-tinyText generation4.5 GB392K
ornith-ai/Ornith-1.5-35B-A3B-MLXMLXText generation69.3 GB344K
ornith-ai/Ornith-1.5-9B-MLX-8bitMLXText generation9.5 GB313K
ornith-ai/Ornith-1.5-9B-MLXMLXText generation17.9 GB313K
mlx-community/Qwen2.5-Coder-7B-Instruct-4bitMLXfrom Qwen/Qwen2.5-Coder-7BText generation4.3 GB259K
ornith-ai/Ornith-1.5-9B-MLX-4bitMLXText generation5.1 GB255K
NexVeridian/Qwen3-Coder-Next-8bitMLXfrom Qwen/Qwen3-Coder-NextText generation84.7 GB225K
ornith-ai/Ornith-1.5-9B-MLX-6bitMLXText generation7.3 GB209K
lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-4bitMLXfrom deepseek-ai/DeepSeek-R1-0528-Qwen3-8BText generation4.6 GB198K
ornith-ai/Ornith-1.5-35B-A3B-MLX-8bitMLXText generation36.9 GB185K
ornith-ai/Ornith-1.5-35B-A3B-MLX-4bitMLXText generation19.5 GB176K
lmstudio-community/DeepSeek-R1-0528-Qwen3-8B-MLX-8bitMLXfrom deepseek-ai/DeepSeek-R1-0528-Qwen3-8BText generation8.7 GB171K
mlx-community/gpt-oss-20b-MXFP4-Q8MLXfrom openai/gpt-oss-20bText generation12.1 GB160K
ornith-ai/Ornith-1.5-35B-A3B-MLX-6bitMLXText generation28.2 GB155K
lmstudio-community/GLM-4.7-Flash-MLX-8bitMLXfrom zai-org/GLM-4.7-FlashText generation31.8 GB154K
lmstudio-community/GLM-4.7-Flash-MLX-6bitMLXfrom zai-org/GLM-4.7-FlashText generation24.4 GB144K
lmstudio-community/LFM2-24B-A2B-MLX-4bitMLXfrom LiquidAI/LFM2-24B-A2BText generation13.4 GB136K
lmstudio-community/LFM2-24B-A2B-MLX-5bitMLXfrom LiquidAI/LFM2-24B-A2BText generation16.4 GB134K
lmstudio-community/LFM2-24B-A2B-MLX-8bitMLXfrom LiquidAI/LFM2-24B-A2BText generation25.3 GB134K
lmstudio-community/LFM2-24B-A2B-MLX-6bitMLXfrom LiquidAI/LFM2-24B-A2BText generation19.4 GB134K
mlx-community/Kimi-K2.5MLXfrom moonshotai/Kimi-K2.5Text generation658 GB110K
lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-4bitMLXfrom Qwen/Qwen3-Coder-30B-A3B-InstructText generation17.2 GB87.5K
Jackrong/MLX-Qwopus3.5-9B-v3-4bitMLXfrom Jackrong/Qwopus3.5-9B-v3Text generation5.1 GB81.2K
lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-5bitMLXfrom Qwen/Qwen3-Coder-30B-A3B-InstructText generation21 GB74.7K
lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-8bitMLXfrom Qwen/Qwen3-Coder-30B-A3B-InstructText generation32.5 GB74.2K
lmstudio-community/Qwen3-Coder-30B-A3B-Instruct-MLX-6bitMLXfrom Qwen/Qwen3-Coder-30B-A3B-InstructText generation24.8 GB73.1K
lmstudio-community/Qwen2.5-Coder-14B-Instruct-MLX-4bitMLXfrom Qwen/Qwen2.5-Coder-14B-InstructText generation8.3 GB72.8K
mlx-community/Qwen3-4B-Instruct-2507-4bitMLXfrom Qwen/Qwen3-4B-Instruct-2507Text generation2.3 GB71.3K
lmstudio-community/LFM2.5-1.2B-Instruct-MLX-8bitMLXfrom LiquidAI/LFM2.5-1.2B-InstructText generation1.2 GB56K
lmstudio-community/Qwen2.5-Coder-14B-Instruct-MLX-8bitMLXfrom Qwen/Qwen2.5-Coder-14B-InstructText generation15.7 GB55.1K
lmstudio-community/LFM2.5-1.2B-Instruct-MLX-4bitMLXfrom LiquidAI/LFM2.5-1.2B-InstructText generation663 MB55.1K
lmstudio-community/LFM2.5-1.2B-Instruct-MLX-6bitMLXfrom LiquidAI/LFM2.5-1.2B-InstructText generation956 MB54.8K
mlx-community/Qwen3-0.6B-8bitMLXfrom Qwen/Qwen3-0.6BText generation649 MB50.8K
mlx-community/Qwen2.5-0.5B-Instruct-4bitMLXfrom Qwen/Qwen2.5-0.5BText generation290 MB43.9K
mlx-community/Qwen3-0.6B-4bitMLXfrom Qwen/Qwen3-0.6BText generation351 MB42.8K
lmstudio-community/Qwen2.5-Coder-32B-Instruct-MLX-4bitMLXfrom Qwen/Qwen2.5-Coder-32B-InstructText generation18.4 GB39K
lmstudio-community/Qwen3-14B-MLX-4bitMLXfrom Qwen/Qwen3-14BText generation8.3 GB38.8K
lmstudio-community/Qwen2.5-Coder-32B-Instruct-MLX-8bitMLXfrom Qwen/Qwen2.5-Coder-32B-InstructText generation34.8 GB35.9K
mlx-community/Qwen3-8B-4bitMLXfrom Qwen/Qwen3-8BText generation4.6 GB33.5K
lmstudio-community/Qwen3-14B-MLX-8bitMLXfrom Qwen/Qwen3-14BText generation15.7 GB33.2K
mlx-community/Llama-3.2-3B-Instruct-4bitMLXfrom mlx-community/Llama-3.2-3B-InstructText generation1.8 GB32.9K
lmstudio-community/Qwen3-4B-Instruct-2507-MLX-4bitMLXfrom Qwen/Qwen3-4B-Instruct-2507Text generation2.3 GB32.8K
mlx-community/Qwen3-Embedding-0.6B-4bit-DWQMLXfrom Qwen/Qwen3-Embedding-0.6BText generation351 MB32.4K
mlx-community/Llama-3.2-1B-Instruct-4bitMLXfrom mlx-community/Llama-3.2-1B-Instruct-bf16Text generation713 MB31.9K
lmstudio-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-4bitMLXfrom nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16Text generation17.8 GB30.6K
lmstudio-community/Qwen3-4B-Instruct-2507-MLX-8bitMLXfrom Qwen/Qwen3-4B-Instruct-2507Text generation4.3 GB30.6K
desert-ant-labs/titleMLXText generation294 MB30.6K
lmstudio-community/Qwen3-4B-Instruct-2507-MLX-6bitMLXfrom Qwen/Qwen3-4B-Instruct-2507Text generation3.3 GB30.4K
lmstudio-community/Qwen3-4B-Instruct-2507-MLX-5bitMLXfrom Qwen/Qwen3-4B-Instruct-2507Text generation2.8 GB30.3K
lmstudio-community/Qwen3-4B-Thinking-2507-MLX-4bitMLXfrom Qwen/Qwen3-4B-Thinking-2507Text generation2.3 GB27.9K
TensorFold/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-MLX-4bitMLXfrom nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16Text generation18.5 GB27.6K
lmstudio-community/Qwen3-4B-Thinking-2507-MLX-8bitMLXfrom Qwen/Qwen3-4B-Thinking-2507Text generation4.3 GB27.4K
lmstudio-community/Qwen3-4B-Thinking-2507-MLX-6bitMLXfrom Qwen/Qwen3-4B-Thinking-2507Text generation3.3 GB27.2K
mlx-community/Qwen3-4B-4bitMLXfrom Qwen/Qwen3-4BText generation2.3 GB27.1K
prism-ml/Ternary-Bonsai-2-27B-mlx-2bitMLXfrom Qwen/Qwen3.8-27BText generation8.6 GB27K
Youssofal/Qwen3.8-27B-MTPLX-Optimized-SpeedMLXfrom Qwen/Qwen3.8-27BText generation20.7 GB27K
lmstudio-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-8bitMLXfrom nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16Text generation33.6 GB27K
lmstudio-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-6bitMLXfrom nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16Text generation25.7 GB26.6K
lmstudio-community/Qwen3-8B-MLX-8bitMLXfrom Qwen/Qwen3-8BText generation8.7 GB26.4K
lmstudio-community/NVIDIA-Nemotron-3-Nano-30B-A3B-MLX-5bitMLXfrom nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16Text generation21.7 GB26.4K
lmstudio-community/Qwen3-8B-MLX-4bitMLXfrom Qwen/Qwen3-8BText generation4.6 GB26.2K
lmstudio-community/Phi-4-mini-reasoning-MLX-4bitMLXfrom microsoft/Phi-4-mini-reasoningText generation2.2 GB25.6K
lmstudio-community/gpt-oss-120b-MLX-8bitMLXfrom openai/gpt-oss-120bText generation124 GB25.4K
mlx-community/gemma-2-9b-it-4bitMLXText generation5.2 GB22.5K
mlx-community/gemma-3-1b-it-qat-4bitMLXfrom google/gemma-3-1b-itText generation772 MB21.3K
lmstudio-community/MiniMax-M2.5-MLX-8bitMLXfrom MiniMaxAI/MiniMax-M2.5Text generation243 GB19.7K
mlx-community/Qwen3-Coder-30B-A3B-Instruct-4bitMLXfrom Qwen/Qwen3-Coder-30B-A3B-InstructText generation17.2 GB18.9K
lmstudio-community/Seed-OSS-36B-Instruct-MLX-8bitMLXfrom ByteDance-Seed/Seed-OSS-36B-InstructText generation38.4 GB18.3K
lmstudio-community/Seed-OSS-36B-Instruct-MLX-4bitMLXfrom ByteDance-Seed/Seed-OSS-36B-InstructText generation20.3 GB17.6K
lmstudio-community/Seed-OSS-36B-Instruct-MLX-5bitMLXfrom ByteDance-Seed/Seed-OSS-36B-InstructText generation24.9 GB17.4K
lmstudio-community/Seed-OSS-36B-Instruct-MLX-6bitMLXfrom ByteDance-Seed/Seed-OSS-36B-InstructText generation29.4 GB17.4K
lmstudio-community/Qwen3-30B-A3B-Instruct-2507-MLX-4bitMLXfrom Qwen/Qwen3-30B-A3B-Instruct-2507Text generation17.2 GB16.6K
lmstudio-community/Qwen3-1.7B-MLX-8bitMLXfrom Qwen/Qwen3-1.7BText generation1.8 GB15.7K
lmstudio-community/Qwen3-1.7B-MLX-4bitMLXfrom Qwen/Qwen3-1.7BText generation984 MB15.7K
mlx-community/Qwen3-1.7B-4bitMLXfrom Qwen/Qwen3-1.7BText generation984 MB15.3K
lmstudio-community/Qwen3-1.7B-MLX-bf16MLXfrom Qwen/Qwen3-1.7BText generation3.5 GB15.3K
lmstudio-community/MiniMax-M2.5-MLX-6bitMLXfrom MiniMaxAI/MiniMax-M2.5Text generation186 GB15.2K
lmstudio-community/MiniMax-M2.5-MLX-4bitMLXfrom MiniMaxAI/MiniMax-M2.5Text generation129 GB15.1K
mlx-community/Qwen3-235B-A22B-Thinking-2507-4bitMLXfrom Qwen/Qwen3-235B-A22B-Thinking-2507Text generation132 GB15.1K
lmstudio-community/Qwen2.5-Coder-7B-Instruct-MLX-4bitMLXfrom Qwen/Qwen2.5-Coder-7B-InstructText generation4.3 GB15.1K
Jundot/DeepSeek-V4-Flash-0731-oQ4e-mtpMLXfrom deepseek-ai/DeepSeek-V4-Flash-0731Text generation166 GB14.9K
lmstudio-community/Qwen3-30B-A3B-Instruct-2507-MLX-8bitMLXfrom Qwen/Qwen3-30B-A3B-Instruct-2507Text generation32.5 GB14.9K
lmstudio-community/Qwen3-30B-A3B-Instruct-2507-MLX-6bitMLXfrom Qwen/Qwen3-30B-A3B-Instruct-2507Text generation24.8 GB14.7K
lmstudio-community/Phi-4-reasoning-plus-MLX-4bitMLXfrom microsoft/Phi-4-reasoning-plusText generation8.3 GB14.3K
RecursiveSI/flint-int4MLXText generation22.6 GB14.2K
mlx-community/Qwen2.5-7B-Instruct-4bitMLXfrom Qwen/Qwen2.5-7BText generation4.3 GB14K
lmstudio-community/LFM2-1.2B-MLX-8bitMLXfrom LiquidAI/LFM2-1.2BText generation1.2 GB13.8K
mlx-community/Qwen1.5-0.5B-Chat-4bitMLXText generation273 MB13.8K
lmstudio-community/LFM2-1.2B-MLX-bf16MLXfrom LiquidAI/LFM2-1.2BText generation2.3 GB13.8K
lmstudio-community/ERNIE-4.5-21B-A3B-MLX-4bitMLXfrom baidu/ERNIE-4.5-21B-A3B-PTText generation12.3 GB13.4K
lmstudio-community/ERNIE-4.5-21B-A3B-MLX-6bitMLXfrom baidu/ERNIE-4.5-21B-A3B-PTText generation17.7 GB13.1K
lmstudio-community/ERNIE-4.5-21B-A3B-MLX-8bitMLXfrom baidu/ERNIE-4.5-21B-A3B-PTText generation23.2 GB13.1K
lmstudio-community/Qwen3-4B-MLX-4bitMLXfrom Qwen/Qwen3-4BText generation2.3 GB13.1K
mlx-community/Qwen2.5-3B-Instruct-4bitMLXfrom Qwen/Qwen2.5-3BText generation1.7 GB12.7K
lmstudio-community/Qwen3-4B-MLX-8bitMLXfrom Qwen/Qwen3-4BText generation4.3 GB12.6K
mlx-community/Qwen2.5-1.5B-Instruct-4bitMLXfrom Qwen/Qwen2.5-1.5BText generation880 MB11.4K
mlx-community/gemma-4-26B-A4B-it-qat-OptiQ-4bitMLXfrom google/gemma-4-26B-A4B-it-qat-q4_0-unquantizedText generation21.9 GB10.9K
mlx-community/Qwen2.5-14B-Instruct-4bitMLXfrom Qwen/Qwen2.5-14BText generation8.3 GB10.3K

The 100 most downloaded are listed here. For the rest, browse them on Hugging Face. See more on Hugging Face

Browse by task

Pages for tasks with at least 5 models, across Core ML, MLX and Apple's own.

More on on-device models

How to run these

  • Core ML: add the .mlpackage to your Xcode project, or download a compiled model when the app first needs it, and load it with the Core ML framework. Some repos ship their own Swift package; the model card says which.
  • MLX: on a Mac with Apple silicon, the mlx-lm and mlx-audio Python packages load these by name; inside an app, MLX Swift runs them on Mac, iPhone and iPad. Mind the memory: in our test a 4-bit 0.5B model peaked at about 0.36 GB on a Mac; larger models need far more.
  • Apple's own: some are Core ML files ready for an app; others are research weights for PyTorch and need converting with coremltools first.

Reading the table

  • Downloads are Hugging Face's count for the last 30 days, not all time. Hugging Face counts requests for a model's main file, so one install can count more than once.
  • Size is the total of the repo's current files on Hugging Face, in the sizes it shows (1 GB = 1,000 MB). Some repos hold several variants of a model, so it can be larger than the one download you need; a dash means it has not been measured yet.
  • Trending is Hugging Face's own score for recent attention. "from" under a name is the model it was made from.
  • Parameters is the parameter count in the model files. A count marked * comes from the model's name (such as 27B) instead: the files gave none, or a much smaller one, as packed low-bit files can.
  • Quantization is read from the model's tags and name, not checked against the files.

Before you use one

The list comes from Hugging Face's public API and is updated daily. We don't test each model. Check the license and read the model card before you ship one in an app: some licenses limit commercial use, and "other" means the terms are on the model card. Models whose names or tags point at adult or safety-removed content are left out, and so are gated models that need a Hugging Face sign-in to download.