Llama 3.2 3B on iPhone

What is Llama 3.2 3B?

Llama 3.2 is Meta’s family of open language models. The 3B version has 3 billion parameters: small enough for a phone, big enough to write, reason and keep context much better than the tiniest models in the library. In Priobs it runs as a 4-bit quantized MLX build on Apple’s GPU.

Who made it

Meta released Llama 3.2 under the Llama 3.2 Community License, which allows on-device apps like Priobs to run it. Meta built the small sizes specifically for phones and laptops, not data centres.

What it’s good at on iPhone

Llama 3.2 3B is great for writing and everyday questions. It beats the sub-1B models at:

• Drafting and editing emails, messages and longer texts
• Following multi-step instructions (“summarise this, then make it three bullet points”)
• Keeping track of a longer back-and-forth
• Brainstorming and explaining ideas clearly

Llama 3.2 3B supports Priobs tools, so it can check your calendar, add reminders or read Health data when you turn those tools on.

It will not out-think a big cloud model on hard technical problems. For the hardest questions in the library, try Qwen 3 4B. For everyday writing, Llama is a strong, fully local choice.

Which iPhones can run it

Llama 3.2 3B needs at least 2.3 GB of memory. Priobs checks your iPhone, hides models it does not have the memory for, and warns you if a model is close to the limit and may be unstable. As a rough guide, an iPhone 12 or later (4 GB of RAM or more) handles it, with more headroom on newer models.