benchgap
Anthropic · model

Claude Haiku 4.5 benchmark scores

As of 2026-11-22, Claude Haiku 4.5 (Anthropic) has measured scores on 18 benchmarks and estimated scores on 96 more.

BenchmarkScoreSource
AA-GPQA Diamond74.4%estimated ± 3.6 pp, high confidence
AA-HLE9.3%estimated ± 6.2 pp, medium confidence
AA-SciCode45.3%estimated ± 3.8 pp, high confidence
Artificial Analysis Intelligence Index12.6%estimated ± 5.8 pp, medium confidence
CritPt0.9%estimated ± 5.3 pp, medium confidence
GDPval-AA45.5%estimated ± 1.7 pp, low confidence
GPQA Diamond80.1%estimated ± 1.8 pp, medium confidence
HLE25.9%estimated ± 9.2 pp, low confidence
AA-Omniscience Accuracy22.2%estimated ± 6.5 pp, medium confidence
SciCode44.5%estimated ± 3.6 pp, high confidence
Terminal-Bench 2.145.5%estimated ± 6.7 pp, low confidence
BrowseComp66.5%estimated ± 6.2 pp, medium confidence
HLE w/ tools53.7%estimated ± 7.5 pp, low confidence
AA Coding Index46.7%estimated ± 5.3 pp, medium confidence
AA-MMMU-Pro64.0%estimated ± 6.1 pp, medium confidence
APEX-Agents-AA19.9%estimated ± 3.6 pp, medium confidence
DeepSearchQA75.2%estimated ± 8.2 pp, low confidence
SWE-bench Verified73.3%measured
CyberGym37.9%estimated ± 11.4 pp, low confidence
JobBench16.0%measured
SWE-bench Pro51.1%estimated ± 4.5 pp, high confidence
WideResearch73.8%estimated ± 1.2 pp, medium confidence
MMLU-Pro81.1%estimated ± 4.8 pp, high confidence
FailureSensorIQ61.8%measured
GPQA75.1%estimated ± 4.5 pp, high confidence
MMLU88.4%estimated ± 1.2 pp, high confidence
Vals GPQA Diamond72.2%measured
Vals LiveCodeBench41.2%measured
Vals MMLU-Pro78.7%measured
Chess Puzzles8.0%measured
AA MMLU-Pro83.5%estimated ± 1.1 pp, medium confidence
Gert Labs33.6%estimated ± 7.7 pp, medium confidence
AA-AnalystAgent31.4%estimated ± 2.7 pp, low confidence
AA EnterpriseOps-Gym40.4%estimated ± 1.2 pp, low confidence
AA Terminal-Bench 2.179.4%estimated ± 2.5 pp, medium confidence
ApprenticeBench0.1%estimated ± 6.0 pp, low confidence
ARC-AGI-149.3%estimated ± 6.5 pp, low confidence
ARC-AGI-20.0%estimated ± 13.8 pp, low confidence
CursorBench 3.140.4%estimated ± 5.8 pp, low confidence
CursorBench 3.261.5%estimated ± 3.3 pp, medium confidence
EBR-bench0.0%estimated ± 9.1 pp, low confidence
FrontierCode 1.1 Main15.7%estimated ± 4.2 pp, low confidence
FrontierSWE v25.5%estimated ± 9.5 pp, low confidence
Furniture Assembly24.7%estimated ± 14.1 pp, low confidence
MaxxBench32.8%estimated ± 10.3 pp, low confidence
MazeBench0.0%measured
Mystery Game Puzzles8.7%estimated ± 2.2 pp, low confidence
NYT Connections14.3%measured
OfficeQA Pro47.1%estimated ± 7.0 pp, low confidence
OSWorld-Verified60.8%estimated ± 8.4 pp, low confidence
ParseBench-149.3%measured
Terminal-Bench 4.16.1%estimated ± 14.9 pp, low confidence
Terminal-Bench Hard44.0%estimated ± 3.9 pp, medium confidence
Vals SWE-bench66.6%measured
Terminal-Bench 2.1 (Vals AI)43.8%measured
VulcanBench v376.2%measured
CWE-Bench-AA44.0%estimated ± 10.6 pp, low confidence
DeepsecBench-AA21.9%estimated ± 10.3 pp, low confidence
Bug Hunt Bench19.0%estimated ± 8.6 pp, low confidence
CursorBench 4.030.8%estimated ± 3.4 pp, medium confidence
CWE-bench v157.4%estimated ± 5.1 pp, medium confidence
DeepSWE56.3%estimated ± 7.2 pp, low confidence
DRACO86.9%estimated ± 0.7 pp, low confidence
FrontierPhysics0.0%estimated ± 6.1 pp, low confidence
HLE w/o tools24.6%estimated ± 4.9 pp, high confidence
OSWorld 2.166.4%estimated ± 6.1 pp, medium confidence
OSWorld 2.1 Offline15.7%measured
PostTrainBench v1.112.5%estimated ± 3.7 pp, medium confidence
SpireClimb0.0%estimated ± 10.3 pp, low confidence
StealthMark30.2%estimated ± 9.7 pp, low confidence
SWE-bench Multimodal0.0%estimated ± 5.6 pp, low confidence
Toolathlon-Verified73.4%estimated ± 2.6 pp, high confidence
AA AutomationBench52.8%estimated ± 12.8 pp, low confidence
GDP.pdf16.4%estimated ± 4.1 pp, low confidence
AA Terminal-Bench 4.00.0%estimated ± 13.7 pp, low confidence
CWE-bench53.9%estimated ± 7.9 pp, low confidence
MKA37.3%estimated ± 3.4 pp, medium confidence
SkillsBench official leaderboard (BenchFlow)30.1%measured
BinaryAudit10.5%measured
CompileBench68.9%measured
OTelBench5.8%measured
Vibe Code Bench0.6%estimated ± 8.4 pp, medium confidence
FACTS Multimodal31.8%estimated ± 5.6 pp, low confidence
Hermes Index37.2%estimated ± 3.1 pp, high confidence
ParseBench-259.8%estimated ± 4.3 pp, high confidence
CharXiv79.8%estimated ± 5.6 pp, low confidence
ExploitBench7.6%estimated ± 13.4 pp, low confidence
ExploitGym11.9%estimated ± 6.3 pp, medium confidence
Claw-Eval54.1%estimated ± 4.5 pp, medium confidence
MCP Atlas38.1%estimated ± 8.6 pp, medium confidence
MMMU-Pro71.0%estimated ± 6.9 pp, medium confidence
NL2Repo38.9%estimated ± 4.7 pp, medium confidence
PostTrainBench v17.8%estimated ± 3.6 pp, medium confidence
ScreenSpot Pro72.6%estimated ± 10.1 pp, low confidence
Terminal-Bench 2.053.4%estimated ± 5.8 pp, medium confidence
Toolathlon29.7%estimated ± 7.4 pp, low confidence
AA LiveCodeBench85.4%estimated ± 2.6 pp, high confidence
React Native Evals76.6%estimated ± 2.7 pp, medium confidence
ResearchClawBench14.9%estimated ± 2.6 pp, low confidence
SWE-Rebench57.0%estimated ± 3.6 pp, medium confidence
PostTrainBench10.3%estimated ± 3.7 pp, medium confidence
DAYJOB0.4%estimated ± 6.3 pp, low confidence
DAYJOB Healthcare0.1%estimated ± 6.1 pp, low confidence
Finance Agent v256.1%estimated ± 3.8 pp, medium confidence
ScarfBench38.2%estimated ± 12.3 pp, low confidence
BlenderWorld (text-only)70.6%estimated ± 4.0 pp, low confidence
Cookie-Bench73.4%estimated ± 2.5 pp, low confidence
LiveCodeBench75.1%estimated ± 14.6 pp, low confidence
Agents' Last Exam35.4%estimated ± 11.0 pp, low confidence
OpenHarmony Bench v1.050.1%estimated ± 3.2 pp, medium confidence
SEC-Bench Pro13.4%estimated ± 12.3 pp, low confidence
LifeVal75.2%estimated ± 3.8 pp, medium confidence
MMMU-Pro w/ Python74.2%estimated ± 2.7 pp, medium confidence
ExploitBench Internal Port0.0%estimated ± 1.4 pp, low confidence