2 ms·
> Thinking blocks from previous assistant turns are preserved in model context by default This seems like a huge change no? I often use max thinking on the ass
by nojs 11mo ago
> Thinking blocks from previous assistant turns are preserved in model context by default
This seems like a huge change no? I often use max thinking on the assumption that the only downside is time, but now there’s also a downside of context pollution
- sothatsit 11mo agoOpus 4.5 seems to think a lot less than other models, so it’s probably not as many tokens as you might think. This would be a disaster for models like GPT-5 high, but for Opus they can probably get away with it.