Given that our Math leaderboard (GSM1K) is now relatively saturated (most models score >90), we are working on producing a new Math benchmark to properly discern between models.

→ View original post on X — @alexandr_wang