PRISM Leaderboard

No published quarters yet

Results are published quarterly, after every submission in the quarter has had its embargo period.

Entries labelled operator baseline were run by the benchmark’s maintainers. They are operator-reported and are not third-party reproducible.
The published evaluator will not run end to end: there is no public sandbox, and there is not going to be one. Runs happen on operator infrastructure, quarterly.
This board is small and mostly operator-run. That is the honest state of it, not a staging artefact.

How this is measured · The benchmark on GitHub