GPT-5.4 - Hackdot

Re:GPT-5.4 by bicx on Thursday March 05, 2026 @02:47PM (#47266371)

That last benchmark seemed like an impressive leg up against Opus until I saw the sneaky footnote that it was actually a Sonnet result. Why even include it then, other than hoping people don't notice?

[ Parent | on HN ]

Re:GPT-5.4 by osti on Thursday March 05, 2026 @03:00PM (#47266514)

It's only that one number that is for sonnet.

[ Parent | on HN ]

Re:GPT-5.4 by 0123456789ABCDE on Thursday March 05, 2026 @03:46PM (#47267051)

except for the webarena-verified

[ Parent | on HN ]

Re:GPT-5.4 by conradkay on Thursday March 05, 2026 @02:57PM (#47266477)

Sonnet was pretty close to (or better than) Opus in a lot of benchmarks, I don't think it's a big deal

[ Parent | on HN ]

Re:GPT-5.4 by jitl on Thursday March 05, 2026 @03:03PM (#47266561)

wat

[ Parent | on HN ]

Re:GPT-5.4 by 0123456789ABCDE on Thursday March 05, 2026 @03:54PM (#47267142)

maybe gp's use of the word "lots" is unwarranted

https://artificialanalysis.ai indicates that sonnect 4.6 beats opus 4.6 on GDPval-AA, Terminal-Bench Hard, AA Long context Reasoning, IFBench.

see: https://artificialanalysis.ai/?models=claude-sonnet-4-6%2Ccl...

[ Parent | on HN ]

Re:GPT-5.4 by conradkay on Thursday March 05, 2026 @10:47PM (#47270622)

I was basing it off my recollection of this:

https://www.anthropic.com/_next/image?url=https%3A%2F%2Fwww-...

basically 9/13 are very close

[ Parent | on HN ]