CodeUpdateArena introduces the first benchmark for evaluating how code-generating AI models can update their knowledge when APIs evolve.

• Tests LLMs' ability to incorporate new programming API information
• Measures knowledge editing effectiveness across different programming languages
• Evaluates both traditional fine-tuning and newer knowledge editing techniques

This research addresses a critical engineering challenge: as software libraries constantly change, AI code assistants must stay current with the latest API specifications to remain reliable development tools.

CodeUpdateArena: Benchmarking Knowledge Editing on API Updates