“On DataCurve's DeepSWE benchmark, GPT-5.4 scored 56%, while Opus 4.7 scored 54%.”