Gemini 4 wins in benchmarks. Why this says less and less about model quality

Google has unveiled Gemini 4 Argon, which reportedly leads in 12 out of 18 benchmarks. However, experts note that this follows the same pattern as the Gemini 3.1 Pro release, which later underperformed against competitors in real-world tasks. The article highlights that synthetic test results are becoming increasingly less representative of the actual capabilities of large language models. Despite impressive figures in performance tables, the gap between 'paper' performance and practical application continues to widen. The author points out that the AI industry is increasingly facing the issue where benchmarks no longer reflect the true quality of neural networks, making them a less reliable tool for users and developers when choosing a model for real-world business tasks.