If the main improvements are coming from improving the harnesses and other things around the AI, I can't help but wonder if the incremental improvement in each model is worth the training costs. I know these companies are in a "race" (where's the finish line, though?), but personally I don't see a huge difference between like Opus 4.7 vs 4.6 vs 4.5, and the newer Codex seems very expensive for the gains it theoretically provides.