Comment by orwin

Honestly, even if anthropic models are better than OpenAIs, I don't understand how they want to make money. In October last year, the frontier models from US companies were so much better than the cheapest models, I thought it was over, but nowadays, even smaller models perform adequately enough. I now use free models (or rather, very cheap ones) in all of my personal projects, and even though anthropic's harness is hard to replace on complex cases (it helps understand where the LLM failed better, which allows to correct the mistakes more easily), I'm pretty sure pure LLM gains are less and less with each new models.