Users praise the CritPt benchmark as most underrated after it ranked GPT-5.6 Sol highest.
Based on 1 visible X reactions from 6 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@scaling01 most underrated benchmark
GPT 5.6 does not advance much on CritPt. There's stagnation at the frontier since GPT 5.4, ≈30% seems like a ceiling, which is very interesting, don't remember when I last saw this. but boy does it destroy Fable on cost. Even Terra is better and over 7 times cheaper. https://x.com/teortaxesTex/status/2075426654197883189/photo/1 https://twitter.com/ArtificialAnlys/status/2075423964378366427
This is a fairly rare benchmark response shape. GPTs are doing vastly better job with less compute, but they don't solve drastically more problems. @htihle @scaling01 what do you think?
CritPt is the final boss of benchmarks https://x.com/scaling01/status/2075280638991638772/photo/1
Users praise the CritPt benchmark as most underrated after it ranked GPT-5.6 Sol highest.
Based on 1 visible X reactions from 6 accounts; directional sample.
Ask a question below.
Published answers will appear here.