Snorkel AI introduces Senior SWE Bench for realistic coding tasks

The benchmark focuses on underspecified feature tasks that resemble real-world software engineering. It aims to evaluate LLMs on complex, multi-step coding with ambiguous requirements.