True Index: Domain Rankings

Private Top-Tier (S & A) standing across 3 core domains
Agentic & Enterprise49.4p
#23
5 Top-Tier private tests
Coding & SWE62.7p
#17
5 Top-Tier private tests
Math & Scientific Reasoning58.5p
#19
3 Top-Tier private tests

Head-to-Head: Clean Evals vs. Legacy Benchmarks

3 matched domains · positive boost means higher legacy scores
Capability Domain Clean Evals (Tiers S & A) Legacy Benchmarks (Tiers C & D) Public Lift (Boost)
Agentic & Enterprise49.4p (5)46.3p (2)-3.1p
Coding & SWE62.7p (5)39.2p (3)-23.6p
Math & Scientific Reasoning58.5p (3)67.8p (3)+9.4p
Clean Evals Avg (Tiers S & A)
55.3p
15 private, leak-resistant tests
Legacy Benchmarks Avg (Tiers C & D)
51.1p
8 public or aging tests
Matched Public Lift (Boost)
-12.7p
Near zero or negative indicates clean generalization
Machine-Readable Export

Audited Benchmark Performance of Claude Haiku 5.5

Grouped by benchmark integrity so private capability signal stands apart from aging public tests.

Top-Tier Validated Tests (Tiers S & A) (15)

Private, leak-resistant tests that build real capability signal.

BenchmarkBenchmark Integrity StatusDomainRankPercentile
MLCR-AAPrivate / FreshMath & Scientific Reasoning#8/4686.0p
FrontierCode-AA 2.0Private / FreshCoding & SWE#14/4672.9p
AA-Briefcase v1.1Private / FreshAgentic & Enterprise#16/5069.4p
Code2Video-AAPrivate / FreshCoding & SWE#7/2366.5p
MKA (Multimodal Knowledge Agent)Private / FreshAgentic & Enterprise#13/2358.8p
FrontierMathPrivate / FreshMath & Scientific Reasoning#16/2750.4p
Harvey LAB-AA v1.1Private / FreshAgentic & Enterprise#19/2137.2p
AutomationBench-AAPrivate / FreshAgentic & Enterprise#48/504.1p
AA-AnalystAgentPrivate / FreshAgentic & Enterprise#6/2477.7p
CyberBench v1.1Private / FreshCoding & SWE#5/2577.7p
LiveBenchPrivate / FreshGeneral Knowledge & QA#17/2854.2p
AA-3DPrivate / FreshCoding & SWE#15/2248.5p
Code Migration v1.1Private / FreshCoding & SWE#13/2448.1p
AA-Omniscience v1.1Private / FreshGeneral Knowledge & QA#26/4747.0p
CritPt v2Private / FreshMath & Scientific Reasoning#23/4039.0p

Mid-Tier Evaluations (Tier B) (2)

Partially exposed or transitional evaluations with moderate leakage risk.

BenchmarkBenchmark Integrity StatusDomainRankPercentile
HLE-DiamondSemi-ExposedGeneral Knowledge & QA#22/4956.2p
VIALS-AASemi-ExposedMath & Scientific Reasoning#12/2350.0p

Legacy & Vulnerable Benchmarks (Tiers C & D) (8)

Public or aging benchmarks prone to overfitting and training-data leaks.

BenchmarkBenchmark Integrity StatusDomainRankPercentile
HealthBenchPublic / AgingMath & Scientific Reasoning#5/1684.2p
BioMysteryBenchPublic / AgingMath & Scientific Reasoning#12/2360.7p
Terminal-Bench-Science 0.1Public / AgingMath & Scientific Reasoning#19/4158.6p
SWE Atlas - Codebase QnAPublic / AgingCoding & SWE#11/1547.5p
GDPval-AA v2.1Public / AgingAgentic & Enterprise#22/4346.8p
GDP.pdfPublic / AgingAgentic & Enterprise#27/5045.9p
Terminal-Bench 4.0Public / AgingCoding & SWE#27/5044.9p
SWE-bench Pro V2Public / AgingCoding & SWE#8/825.1p