MATH leaderboard
MATH
1 models tested · Updated 2026-04-23 · Verified sources only
Hunyuan Hy3 Preview Base leads at 76.28%
1
Hunyuan Hy3 Preview Base
Tencent · GitHub/Tencent-Hunyuan-Hy3-preview · 2026-04-23
Beats DeepSeek-V3 base (59.37) and GLM-4.5 base (61.0) decisively.
76.28%
benchmark.space canonical public ledger of AI capability