Is the 20B model small enough to be used on iPhones? Can anyone give some insights?
OpenAI (@OpenAI) profile pictureOpenAI
@OpenAI5 Aug 2025
Our open models are here. Both of them. openai.com/open-models
Open models by OpenAI

Open models by OpenAI

Advanced open-weight reasoning models to customize for any use case and run anywhere.

openai.com

Aug 6, 2025 · 7:14 AM UTC

8
10
7,339
RelevantRecentLikes
No
2
365
Nope. Even if not all 20B parameters are active. The best iPhones currently have a max of 8GB of RAM. During my testing with my app, 8B 5GB models reach the limits. They run slow, get extremely hot, and there are high chances of iOS killing the app. Maybe an iPhone with 16Gb of RAM like the 1-2To iPads with M chips.
2
175
It’s quantized to 4 bits with floating-point precision (MXFP4), which occupies 12.11GB. You need a new quantized model with reduced precision, ideally 2 bits or less.
3
282
Last year I did some laboration with this one on iOS: huggingface.co/microsoft/Phi… For the purpose of the test it worked fine on my iPhone 14 (no pro). It was kind of slow, but worked.
microsoft/Phi-3-mini-4k-instruct-onnx · Hugging Face

microsoft/Phi-3-mini-4k-instruct-onnx · Hugging Face

Text Generation model. Download, explore files and run it on Hugging Face.

huggingface.co
2
107
No I don’t think so. Unless you have a high end iPhone
1
183
No way, sadly.
1
27
I guess maybe on some androids but probably not on any iphone 12gb max
160
What size are the models they are offering I haven’t checked it out yet but seems exciting possibilities
245