The author outlines their current AI stack, emphasizing the importance of using multiple models to optimize performance. For general inquiries and research, they primarily rely on GPT 5.2 Thinking and Pro models. They find GPT Pro particularly effective for thorough research due to its ability to pull information from multiple sources. In contrast, non-Thinking versions of GPT feel less reliable. The author mentions Claude 4.5 Opus for coding questions and feedback, noting its refreshing tone and adequate speed, while Gemini 3 Pro is reserved for concept explanations and multimodal tasks. They feel that Gemini’s search capabilities lag behind GPT’s, especially for recent information. The article emphasizes the necessity of switching between models to tackle complex tasks. The author frequently finds that a peer model can resolve issues where one model gets stuck. They argue that if each model had a low probability of success, switching would rarely yield results. However, the current landscape shows that models have a reasonably high success rate for various tasks, making this multi-model approach effective. The author appreciates the value derived from their AI subscriptions, highlighting how cutting-edge models remain expensive but essential for advanced tasks. The distinction between frontier models and open models is also discussed, with the author noting that open models are significantly cheaper but often lag behind in performance. They predict that the gap in quality and reliability will remain, as open models have yet to reach the capabilities of models like Claude 4.5 Opus or GPT 5.2 Thinking. As 2025 introduced various new AI tools and agents, the author stresses the importance of exploring different providers rather than sticking to one. The rapidly evolving nature of AI requires continual experimentation with new tools to stay ahead.