Re:MaxxBench – "Nice DeepSWE score. But is it true?"
by rapind on Saturday November 07, 2026 @04:38PM
(#50408295)
Different expectations maybe. In my experience, DSF 4.1 high/max is basically what Terra should be for coding, and is very acceptable (Terra is hot garbage, yes even the latest 6.1). Luna at xhigh+ stands out as a cheap implementor/reviewer of very very well defined tasks, but goes off the rails if you're not very specific, Sol and Sonnet at medium are basically powerhouses for implementation with a half decent plan, and Opus 5.5 high is an excellent planner (and implementor).
[ Parent
| on HN ]
There are no comments on this comment yet.