OpenAI and Anthropic have recently released their latest coding assistant models, GPT-5.3-Codex and Claude Opus 4.6. OpenAI's Codex 5.3 shows significant improvement over its predecessor, particularly in speed and versatility across tasks like data analysis and git operations. Users report that Codex 5.3 operates more like Claude, with faster feedback and a better fit for varied tasks. However, while Codex 5.3 has gained ground, Opus 4.6 remains superior in usability, providing a smoother experience for broader applications. The article notes that the standard benchmarking methods for evaluating these models are becoming less relevant. Traditional metrics that once highlighted performance gains have diminished in significance as user experience takes precedence. The author emphasizes the need for users to engage with multiple models and adapt their skills to manage these AI agents effectively. Despite Codex 5.3's advancements, it still lags behind Claude in real-world usability for non-expert users, suggesting a gap that OpenAI needs to address. Anthropic's approach, particularly with Claude 4, has set it apart by prioritizing user experience over traditional metrics. The article reflects on a shift in the industry where the focus is moving towards practical application and user feedback rather than just performance scores. This shift is underscored by the assertion that AI's progress might not always align with benchmark improvements, urging a more nuanced understanding of advancements in AI capabilities.