The one saving grace I see here is that open models are getting really good, and they're already profitable at an affordable price.

So maybe it's true you won't get "the best", but I don't think you'll be that far off.

There's this weird race where I have in my head some level of LLM performance which is "good enough" and the open models keep improving to that level, but by the time that they do my "good enough" has acclimatized to what I'm used to doing with the latest frontier models and what the open models are isn't good enough anymore.

The "good enough" points so far have been

- "as good as ChatGPT"

- "as good as GPT4"

- "as good as Sonnet 3.5"

- "as good as Opus 4.5 or Codex 5.2"

Anyway, we'll see where the chinese models are in a year, and we'll see where my expectations are. Hopefully they overlap at some point.