DeepSWE Benchmark Results for GPT 5.6

I wouldn't call it borderline. If we take the chart seriously (if), it means for every budget, the best performance is achieved by a 5.6 variant.