This is a useful but qualified benchmark comparison: it finds no statistically significant scaffold advantage for either model, while leaving room for token-budget and evaluation-design effects.
Limits of the evidence: The reported differences are not statistically significant; METR notes that the specialized scaffolds may have needed larger token budgets. The source is a single report and the supplied text is truncated.
What to watch
Taken from the first report; the whole story has not been re-assessed yet.
Measuring Time Horizon using Claude Code and Codex
METR BlogFeb 13, 2026First report
No revisions yet.
No maturity ladder applies to this desk.
Sens.ai aggregates and assesses published reporting. The assessment above is machine generated; the original sources are authoritative.