Model Picker

Short version

Model	Best for	Start here if	Avoid it if
E2B	Phones and low-memory devices	Your main goal is “make it run locally today”	You expect frontier-level depth
E4B	Better phones, tablets, light desktops	You want better quality than E2B without a workstation	Your device already struggles with small models
26B A4B	Macs and stronger desktops	You have real memory headroom and want the best tradeoff	You are memory constrained
31B	Workstations	Quality matters most and your hardware is serious	You want fast setup or broad device compatibility

Picking by intent

• Offline chat, translation, note drafting, quick Q&A on mobile: E2B first, E4B if the device can handle it.
• General desktop use and experimentation: E4B or 26B A4B.
• Local research assistant, stronger writing, higher-quality reasoning: 26B A4B.
• “I am okay trading convenience for quality”: 31B.

Picking by hardware reality

• If you are not sure the model will fit, choose the smaller option.
• If load time already feels painful, the next larger model is usually the wrong move.
• If you need a dependable phone demo, choose stability over benchmark wins.

Common mistake

The most common failure pattern is:

1. See a benchmark for 31B.
2. Download 31B onto the wrong machine.
3. Blame the runtime.

Usually the problem is not the runtime. It is the fit between model size and available memory.

What to do next

• Need phone help? Go to Android or iPhone and iPad.
• Need desktop help? Go to Mac or Windows.
• Already failing? Open Out of Memory.