FrontierCode
A coding benchmark from Cognition that grades the quality of production code rather than whether it merely runs.
FrontierCode is a benchmark built by Cognition that judges code the way a colleague reviewing a pull request would: not just does it work, but is it the kind of code you would want in a real project. That is a higher bar than the older tests, which mostly checked whether a small function produced the right output.
It shows up constantly in model announcements because the difference between “produces working code” and “produces code worth keeping” is exactly where the current competition sits. Scores in the forties are normal for strong models, which is a useful reality check against claims that AI has finished the job of writing software.