3 ms·
Yeah, GPT 5.5 + Fable beating either individually is belivable, but 2x Opus > Fable is what makes me a bit dubious about the whole thing. They might be measurin
by qsort 4mo ago
Yeah, GPT 5.5 + Fable beating either individually is belivable, but 2x Opus > Fable is what makes me a bit dubious about the whole thing. They might be measuring skills that are too specific or benefit a lot from more tokens being thrown at them. Also Claude Code (the harness) is not the best at the moment, that might be part of it as well?
- andai 4mo agoWhat throws me off is DeepSeek beating both Opus 4.8 and GPT 5.5. That definitely doesn't sound right.