ResearchBenchmarksunsound inference on Hiring Assistant (test)Follow0.968KappaTextGrad0.885840.907170.92850.94983May 25, 2026Evaluation ResultsMethodMethodLinksKappaTextGradN (best-of-N)=1, 3, Ag...N (best-of-N)=1, 3, Agent Model=GPT-4o2026.050.968SPEARN (best-of-N)=3, 5, Ag...N (best-of-N)=3, 5, Agent Model=GPT-5.42026.050.937GEPAN (best-of-N)=1, 3, Ag...N (best-of-N)=1, 3, Agent Model=GPT-5.42026.050.889