Model library
The Priobs model library: 10 on-device AI models.
Priobs offers 10 downloadable open models from Alibaba, Google, Meta, Microsoft and Liquid AI, plus Apple’s built-in Foundation Model. Every download is a 4-bit quantized MLX build that runs on your iPhone. No server, no Priobs account, and no connection needed once it is installed.
Priobs checks how much memory your iPhone has, hides models that will not fit, and warns you about ones that may be unstable. It also recommends a model for your device.
10 downloadable models
Open models, running on your hardware.
Sorted by download size, smallest first. Qwen 3 4B is the most capable. Gemma 4 E2B is the newest from Google and the biggest download.
Qwen 2.5 0.5B
Alibaba
The smallest model in the library and by far the quickest. Good for short questions, quick rewrites and testing what on-device AI feels like on older iPhones.
- Download
- 288 MB
- Min RAM
- 512 MB
- License
- Apache 2.0
LFM2 1.2B
Liquid AI
A compact model designed for efficient edge-device inference. It is a strong choice for quick everyday prompts and supports tools in Priobs.
- Download
- 663 MB
- Min RAM
- 862 MB
- License
- LFM Open License v1.0
Gemma 3 1B
Google’s compact open model. A sensible default if you want noticeably better answers than the 0.5B tier without a long download.
- Download
- 767 MB
- Min RAM
- 1 GB
- License
- Gemma License
Qwen 2.5 1.5B
Alibaba
Trained on a broad multilingual mix. The one to pick if you chat, write or translate in more than one language.
- Download
- 878 MB
- Min RAM
- 1.2 GB
- License
- Apache 2.0
Qwen 3 1.7B
Alibaba
A compact reasoning model that can work through a problem before answering. It also supports Priobs tools while staying below a 1 GB download.
- Download
- 982 MB
- Min RAM
- 1.2 GB
- License
- Apache 2.0
Llama 3.2 3B
Meta
Meta’s 3B instruction-tuned model. Good at longer writing, editing and everyday questions, and it supports Priobs tools.
- Download
- 1.8 GB
- Min RAM
- 2.3 GB
- License
- Llama 3.2 License
Phi-4 Mini 3.8B
Microsoft
Microsoft’s Phi-4 Mini is strong at maths, logic and coding help for its size. MIT licensed, with tools in beta.
- Download
- 2.2 GB
- Min RAM
- 2.8 GB
- License
- MIT
Qwen 3 4B
Alibaba
The most capable model in the library. It thinks before it answers, supports Priobs tools, and handles the hardest everyday tasks you can run on an iPhone.
- Download
- 2.3 GB
- Min RAM
- 2.9 GB
- License
- Apache 2.0
Qwen 3 4B Thinking
Alibaba
A reasoning-focused Qwen 3 variant for harder multi-step work. It takes longer to answer and exposes experimental tool support in Priobs.
- Download
- 2.3 GB
- Min RAM
- 2.9 GB
- License
- Apache 2.0
Gemma 4 E2B
Google’s newest open model and the biggest download in the library. Strong reasoning, with tools in beta. It needs a recent iPhone with plenty of memory.
- Download
- 3.6 GB
- Min RAM
- 4.5 GB
- License
- Gemma License
Side by side
Full specifications
Numbers come straight from the model catalogue inside the app.
- Qwen 2.5 0.5BAlibaba
Fastest responses, good for simple tasks
- Download
- 288 MB
- Min RAM
- 512 MB
- Tools
- No
- License
- Apache 2.0
- LFM2 1.2BLiquid AI
Built for speed on phones, snappy short replies
- Download
- 663 MB
- Min RAM
- 862 MB
- Tools
- Tools
- License
- LFM Open License v1.0
- Gemma 3 1BGoogle
Balanced speed and quality
- Download
- 767 MB
- Min RAM
- 1 GB
- Tools
- No
- License
- Gemma License
- Qwen 2.5 1.5BAlibaba
Strong multilingual support
- Download
- 878 MB
- Min RAM
- 1.2 GB
- Tools
- No
- License
- Apache 2.0
- Qwen 3 1.7BAlibaba
Compact model that thinks before answering
- Download
- 982 MB
- Min RAM
- 1.2 GB
- Tools
- Tools
- License
- Apache 2.0
- Llama 3.2 3BMeta
Great for writing and everyday questions
- Download
- 1.8 GB
- Min RAM
- 2.3 GB
- Tools
- Tools
- License
- Llama 3.2 License
- Phi-4 Mini 3.8BMicrosoft
Strong at math and step-by-step answers
- Download
- 2.2 GB
- Min RAM
- 2.8 GB
- Tools
- Tools (beta)
- License
- MIT
- Qwen 3 4BAlibaba
Most capable, best for complex tasks
- Download
- 2.3 GB
- Min RAM
- 2.9 GB
- Tools
- Tools
- License
- Apache 2.0
- Qwen 3 4B ThinkingAlibaba
Works through problems step by step, slower to answer
- Download
- 2.3 GB
- Min RAM
- 2.9 GB
- Tools
- Tools (beta)
- License
- Apache 2.0
- Gemma 4 E2BGoogle
Google’s latest with strong reasoning
- Download
- 3.6 GB
- Min RAM
- 4.5 GB
- Tools
- Tools (beta)
- License
- Gemma License
- Apple IntelligenceApple
Built into your iPhone, nothing to download
- Download
- None
- Min RAM
- Built in
- Tools
- Tools
- License
- Apple system model
Which one should I pick?
Start small. Upgrade when you feel the ceiling.
Bigger models give better answers but need a longer download, more storage and more memory. There is no wrong first choice. Keep several installed and switch any time.
Older or entry-level iPhones
Qwen 2.5 0.5B
288 MB download, fastest replies, best for short questions and quick rewrites.
Most current iPhones
Llama 3.2 3B
1.8 GB download. Great for writing and everyday questions, with tool support.
Recent iPhones with more memory
Qwen 3 4B
The most capable model in the library. 2.3 GB download, 2.9 GB of RAM, thinks before it answers and supports tools.