SWE-bench Verified Update

Warp has significantly improved its single-agent coding system, which is powered by GPT-5, to achieve a higher score on the SWE-bench Verified benchmark. The agent now automatically creates and adapts dynamic task lists, allowing it to break down complex problems more effectively than with previous rigid plans. To better handle long conversations, the system intelligently summarizes context and optimizes file edits by returning only the modified code sections. These enhancements, along with new debugging instructions, show that improving a single agent's quality and reliability is a powerful way to boost real-world coding performance.