Comment by lukebechtel

Some people seem to be suggesting these are coverups for quantization...

Those who work on agent harnesses for a living realize how sensitive models can be to even minor changes in the prompt.

I would not suspect quantization before I would suspect harness changes.