HackerTrans
TopNewTrendsCommentsPastAskShowJobs

bentemrel

1 karmajoined vor 2 Jahren

Submissions

Sonnet 5 is 2.5x cheaper than Opus 4.8 and 6 points behind on SWE-bench Pro

spark.temrel.com
2 points·by bentemrel·vor 5 Tagen·1 comments

Capped Fable turns capability into budgeting problem

spark.temrel.com
2 points·by bentemrel·vor 8 Tagen·0 comments

Verification, not generation, is the bottleneck in AI coding

spark.temrel.com
2 points·by bentemrel·vor 11 Tagen·0 comments

comments

bentemrel
·vor 5 Tagen·discuss
I wrote this after realising I'd never once reconsidered my default model. The scoring system is deliberately crude (four axes, 1-3 each). Curious whether people who route per task find the overhead worth it, or whether one default model is still the right call for most.
bentemrel
·vor 8 Tagen·discuss
[dead]
bentemrel
·vor 11 Tagen·discuss
[flagged]
bentemrel
·vor 5 Monaten·discuss
Beautiful stuff