Medium reasoning has regressed since 4.6. While None and Max have improved since 4.6 in our benchmark. We suspect that this is how Claude tries to cope with the

Medium reasoning has regressed since 4.6. While None and Max have improved since 4.6 in our benchmark. We suspect that this is how Claude tries to cope with the increased user base. Note, Google and OpenAI probably did something similar long ago.