Comment by kuukyo

They don't report the pass@1 success rate. They sample multiple solutions from Opus/Fable and CLM decides which one to submit, that's why they get >80%.