benchgap
Anthropic · model

Claude 3 Haiku benchmark scores

As of 2026-11-22, Claude 3 Haiku (Anthropic) has measured scores on 9 benchmarks and estimated scores on 44 more.

BenchmarkScoreSource
AA-GPQA Diamond37.4%measured
AA-HLE4.1%measured
Artificial Analysis Intelligence Index5.6%measured
CritPt0.0%measured
GDPval-AA0.0%estimated ± 11.0 pp, low confidence
GPQA Diamond68.7%estimated ± 2.4 pp, medium confidence
HLE16.7%estimated ± 8.5 pp, low confidence
IFBench53.9%estimated ± 6.6 pp, low confidence
AA-LCR27.7%measured
AA-Omniscience Accuracy17.6%measured
BrowseComp27.2%estimated ± 13.0 pp, low confidence
IFEval87.4%estimated ± 2.8 pp, high confidence
LongBench v251.2%estimated ± 5.5 pp, low confidence
τ²-bench results21.1%measured
AA-MMMU-Pro30.8%measured
BioMysteryBench (human-difficult)0.0%estimated ± 6.8 pp, low confidence
BFCL v420.9%estimated ± 8.5 pp, low confidence
AA Agentic Index0.0%estimated ± 12.7 pp, low confidence
MMLU-Pro81.9%estimated ± 5.0 pp, medium confidence
AA-IFBench36.1%measured
GPQA42.9%estimated ± 6.1 pp, medium confidence
MMLU80.8%estimated ± 0.9 pp, high confidence
Vals GPQA Diamond36.6%estimated ± 3.2 pp, medium confidence
Vals MMLU-Pro75.5%estimated ± 2.9 pp, high confidence
MMMLU83.3%estimated ± 0.7 pp, medium confidence
MMMU45.9%estimated ± 2.7 pp, medium confidence
AA MMLU-Pro83.7%estimated ± 0.8 pp, medium confidence
ARC-AGI-174.7%estimated ± 7.8 pp, low confidence
ARC-AGI-230.5%estimated ± 14.4 pp, low confidence
ParseBench-140.3%estimated ± 6.2 pp, low confidence
HealthBench Professional (raw, Artificial Analysis)11.3%estimated ± 2.5 pp, low confidence
HLE w/o tools20.3%estimated ± 4.9 pp, high confidence
ARC-AGI-31.3%estimated ± 9.8 pp, low confidence
ARC-AGI-3 Provider Adapter2.7%estimated ± 7.3 pp, low confidence
ParseBench-219.6%estimated ± 6.2 pp, low confidence
CharXiv43.6%estimated ± 6.3 pp, low confidence
AI2D_TEST88.0%estimated ± 2.0 pp, low confidence
C-Eval87.2%estimated ± 1.0 pp, low confidence
MMLU-Redux93.2%estimated ± 0.8 pp, low confidence
MMMU-Pro31.0%estimated ± 1.5 pp, medium confidence
ScreenSpot Pro39.2%estimated ± 10.8 pp, low confidence
SuperGPQA64.5%estimated ± 9.3 pp, low confidence
Video-MME (w/o subtitle)69.6%estimated ± 1.7 pp, low confidence
HealthBench Hard0.0%estimated ± 8.6 pp, low confidence
MMLU-Pro (Arcee)77.1%estimated ± 4.0 pp, low confidence
BioMysteryBench (human-solvable)81.7%estimated ± 0.6 pp, low confidence
HealthBench (length-adjusted)54.1%estimated ± 3.7 pp, medium confidence
HealthBench Professional56.7%estimated ± 4.2 pp, low confidence
STEMBench2.3%estimated ± 0.2 pp, low confidence
SimpleQA14.8%estimated ± 4.3 pp, medium confidence
MRCRv222.8%estimated ± 11.3 pp, low confidence
MMLongBench-Doc55.6%estimated ± 2.6 pp, low confidence
BioBasicsBench34.1%estimated ± 3.8 pp, low confidence