MMLU leaderboard
MMLU
3 models tested · Updated 2026-04-29 · Verified sources only
Granite 4.1 30B leads at 80.16%
1
Granite 4.1 30B
IBM · HuggingFace/IBM Granite · 2026-04-29
5-shot MMLU at 30B.
80.16%
2
Granite 4.1 8B
IBM · HuggingFace/IBM Granite · 2026-04-29
5-shot MMLU at 8B.
73.84%
3
Granite 4.1 3B
IBM · HuggingFace/IBM Granite · 2026-04-29
5-shot MMLU; 3B dense baseline.
67.02%
benchmark.space canonical public ledger of AI capability