I keep coming back to this @ArtificialAnlys chart of the dotted Pareto line shows the models offering the most benchmark performance at each cost level, but the cheapest model on that line isn’t automatically the right one for every task. For high volume, repeatable work, I’d start with something like @OpenAI GPT-6 Luna and see if it clears the accuracy bar. GPT-6.1 Sol looks like a compelling middle ground, but for the harder steps where a better answer changes the outcome paying up for @claudeai Opus 5.5 can make sense. On this chart Opus costs roughly 8x as much per task as 6.1 Sol. You want to keep thinking about which steps actually earn that extra spend because as agents make more calls getting that routing decision right could matter a lot. ---> continuing in a second post