BenchPress predicts a score of 70.3 for Kimi-K3 on ARC-AGI-2 This would put it between Opus 4.6 (120k, high) and GPT-5.4-xhigh
@scaling01 I think 60 China bros aren't ARC-maxxing would be nice to be wrong though
@teortaxesTex I think it's likely above 60% but no higher than 75%
Some users sarcastically dismissed the BenchPress prediction of Kimi-K3 scoring 70.3 on ARC-AGI-2 as unscientific because of the imprecise wide confidence range cited.
Based on 1 visible X reactions from 6 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@scaling01 92% trust with a 20 point range, cool cool very precise very scientific
BenchPress predicts a score of 70.3 for Kimi-K3 on ARC-AGI-2 This would put it between Opus 4.6 (120k, high) and GPT-5.4-xhigh
@scaling01 I think 60 China bros aren't ARC-maxxing would be nice to be wrong though
@teortaxesTex I think it's likely above 60% but no higher than 75%
Some users sarcastically dismissed the BenchPress prediction of Kimi-K3 scoring 70.3 on ARC-AGI-2 as unscientific because of the imprecise wide confidence range cited.
Based on 1 visible X reactions from 6 accounts; directional sample.
Ask a question below.
Published answers will appear here.
