SWE-Bench Failures: When Coding Agents Spiral into 693 Lines of Hallucinations(www.surgehq.ai)

20 pointsby landonxi12 hours ago1 comment

egillie12 hours ago
Is this because GPT-5 hallucinates less in general?