[SyzygyRhythm]

As others said, you can run more than one at a time, but even singly, 24 GB can be very useful in a hybrid approach. For instance, I recently had to process hundreds of high-res blueprints. Used Fable for the main orchestration, but all of the OCR and other work I did with a small local model. Didn't waste expensive Fable tokens on simple grunt work.

[0x6c6f6c]

What are you using for OCR locally? This is something I've been considering trying for some time with old journals.