Hacker News new | past | comments | ask | show | jobs | submit
Not going to speculate on how they'll do vs. OpenAI, but to your point on a lot of problems now effectively being "solved", I tend to agree.

I have one recent use case now where using Claude to review it, it first wanted to use Fable, got all excited about possible savings from lower defect rates, had to backtrack and admit that Fable provided no benefit whatsoever to justify the extra cost over Opus after it ran evals (in fact, arguably the quality was worse, but that part was a subjective measurement).

There's a mounting number of tasks like this where the value is going to be falling costs whether by switching providers, or moving down their model ladder as the one below gets good enough.

Given Opus 5.5 is now good enough for the task in my example, it means we have every incentive to start evaluating cheaper models now and again, instead of waiting eagerly for new, smarter models.