Reading today's open-closed performance gap

Open models continue to decline compared to closed models, with the AI analytical index being a key benchmark for measuring this gap. This index includes about 10 sub-evaluations reflecting the capabilities of current language models. However, relying on a single number to assess performance overlooks the complex dynamics affecting these models. Measurement standards change periodically, evolving over time and becoming less relevant to real-world applications. For instance, the Gemini 3 model performs excellently in benchmarks but lacks practical relevance in current applications.