Zenith: sota harness for normal models to beat Fable on FrontierSWE

Zenith is a state-of-the-art harness that enables standard, non-specialized language models to outperform the Fable system on the FrontierSWE benchmark by optimizing their software engineering task execution. It is designed for AI researchers and developers seeking to maximize the coding and debugging capabilities of accessible models without requiring custom architectures. The project is interesting because it challenges the assumption that only elite, proprietary models can excel at complex software engineering, potentially democratizing high-performance AI for code.