@opencode That’s true!! I’ve created a script to measure the cost. Still analyzing the results
I have highest subscription (20x) and the hourly limits bings pretty fast.
@NousResearch We are working on benching various combos of open source models to see if we can get Opus levels with much cheaper models as well, stay tuned!
Environment, precision, wrappers, and hardware do matter, that’s basic inference stack 101.
But claiming they erase the massive scaling advantage of a much larger, better-trained frontier model running locally on Mac Studio is pure cope. With proper quantization and MLX, the capability gap remains decisive. Model scale defines the ceiling. Hardware just tunes the knobs.