
@panda_liyin @adalengineer It is working fine for me most of the times. Yes is heavier consumer since the tokenizer was changed. Same characters use more tokens. In some cases it fixes what gpt5.5 doesn't and the other way around, so likely will keep using both.
English

























