Is Claude OK?

Inspired by this Anthropic postmortem, we created this benchmark to see if Claude gets "dumber" over time. We measure the intelligence daily using the open source questions from SoTA benchmarks and highlight whenever the model's performance differs from the established baseline.

Claude Sonnet 4.5 through Anthropic API. That's the one most people reach for when they say "Claude", so that's what this page answers for. If the tracked model changes, we'll say so on the page and establish a new baseline.

We compare today's results to the established baseline. Green means Claude responds as expected. Yellow means noticeable, but minor change in quality of the answers. Red means there's a significant difference in how Claude answers the same set of questions

One dot per day. Hover shows date and how much exactly today differs from baseline. Green: within baseline. Yellow: minor difference. Red: significant difference. A hollow dot means we're yet to run the benchmark today.

To establish the baseline, we track how the model answers the same set of questions over a week. After that, we compare next results to this baseline. We run the same set of questions daily so that we can tell when the model performs differently. Our goal is to track consistency, we do not measure the general intelligence.

Not by default, we keep those details private to minimize noise. We share composite scores, grading confidence, and other technical bits on a need-to-know basis. If you need to know, please DM us on Twitter.

No.