28 ms·
I'm a bit disappointed that Opus 4.6 wasn't in this because the tokenizer changed quite a bit from 4.7 onward. I was so annoyed by 4.7 that I've been forcing 4
by fouc 4mo ago
I'm a bit disappointed that Opus 4.6 wasn't in this because the tokenizer changed quite a bit from 4.7 onward. I was so annoyed by 4.7 that I've been forcing 4.6 ever since. I've been annoyed by 4.8 a bit too, so I haven't felt the urge to move on.
- swyx 4mo agoshared older model numbers here https://www.latent.space/p/ainews-frontiercode-benchmarking https://www.latent.space/p/ainews-frontiercode-benchmarking tldr theres been broad progress despite your observed regressions