pull down to refresh

Tried various Claude models for help with some academic work today. Tried Opus 4.8 High, Opus 4.6 Medium, and Sonnet 5 Medium.

I definitely think 4.6 is still better than 4.8. 4.8 seems to want to get too deep into things and ends up confusing itself / not explaining stuff properly.

I also found some regressions in 4.8 high, where it was trying too hard to apply some format, and give me inferior results, even hallucinate way more often as a result. That's when I learned that 4.8 high is nerf'd and 4.6/4.7 high equals 4.8 xhigh. Switching stopped it from trying to framework things and just do what I say.

reply