Even if different models look similar on paper (and sometimes even outperform others in benchmarks), in real-world use there is often a noticeable quality gap. My impression is that non major clients setups are often missing something - especially in workflow smoothness, fit for programming tasks, and overall user experience.
One important reason is that, in integrated ecosystems, the prompts and interaction patterns used by the client are better aligned with the model. The client is designed together with the model, so the prompting logic, tool usage, and response flow are often better tuned to how that specific model performs in practice.
That is why it makes sense to choose solutions built by companies that develop the entire ecosystem around the model (e.g., Anthropic, OpenAI, Google) and work on their own client / environment. In those cases, the model is not just “strong” in isolation, but also better tuned to an actual way of working.
Practical takeaway: instead of jumping between many “side models” and comparing benchmarks, it is better to commit to one specific company / ecosystem and build your workflow around it. The debate about who currently has the edge (Anthropic vs OpenAI vs Google) is secondary to the bigger point: the advantage comes from polished integration, not only from the model itself.