Critics Flag Errors in Claude Opus 5 Chart
AI researchers highlight errors and misleading scales in Claude's performance table.
Anthropic's official Claude account posted a benchmark table claiming Opus 5 leads coding and knowledge evaluations. Multiple AI engineers and researchers replied with examples of chart errors, including selective highlighting of top scores and y-axis ranges set just above results. Miles Brundage framed the issue as over-reliance on Claude outputs rather than deliberate hype, while others noted repeated visual inaccuracies. The exchange underscores growing scrutiny of how frontier labs present model performance data. No corrections from Anthropic appear in the packet.
Combined views
11.4K
12 posts, first seen 2h ago
Reactions from ranked influencers
12 posts
Multi billion dollar company with the smartest ai in the world can’t make a valid chart
On several coding and knowledge work evaluations, Opus 5 is the new state-of-the-art:
@benhylak AI math
The way to understand the latest Anthropic chart crime is "they trust Claude too much," not that they are trying to hype things Not a big deal in isolation, but the over-trust thing is increasingly dangerous in other contexts
@Miles_Brundage claude said its fine
@Miles_Brundage which chart crime?
Multi billion dollar company with the smartest ai in the world can’t make a valid chart https://twitter.com/claudeai/status/2080699497064083942