ResearchBenchmarksSoftware Task Solving on SWE-bench verifiedFollow1.64Succ/Mtok (All)AHE1.07841.22421.371.5158Apr 28, 2026Evaluation ResultsMethodMethodLinksSucc/Mtok (All)Succ/Mtok (django)Succ/Mtok (sympy)Succ/Mtok (sphinx-doc)Succ/Mtok (matplotlib)Succ/Mtok (scikit-learn)Succ/Mtok (pydata)Succ/Mtok (astropy)AHE2026.041.641.671.481.071.883.42.151.81NexAU02026.041.431.51.430.931.513.0620.82TF-GRPO2026.041.271.351.190.781.332.481.51.26ACE2026.041.11.121.150.621.142.081.371.08