IOL-AI Challenge Benchmarks Linguistic Reasoning via International Linguistics Olympiad Problems

The IOL-AI Challenge evaluates models on discovering hidden linguistic systems under a strict 30-minute T4 compute budget. While Claude Opus 4.8 achieved gold-medal equivalent scores, 14B parameter submissions outperformed larger models, suggesting reasoning capability is not strictly scale-dependent.