Princeton's Kilian Lieret says multi-agent setups solve ProgramBench tasks 2x faster but deliver lower-quality code than single agents

— This data comes from the Opus 4.8 system card.

Original post

OP#72@OFIRPRESSOPKLKilian Lieret|@KLIERET

Very interesting study from Opus 4.8 card: Multi-agents do not deliver better results on ProgramBench, but they get to mediocre solutions 2x faster.

2:29 PM · May 28, 2026 View on X

QUOTE POSTCB#999Cody Blakeney|@CODE_STAR

Something about 9 months and a baby

KLKilian Lieret@KLieret

Very interesting study from Opus 4.8 card: Multi-agents do not deliver better results on ProgramBench, but they get to mediocre solutions 2x faster.

9:29 PM · May 28, 2026 · 5.9K Views

6:07 PM · May 29, 2026 · 340 Views