Benchmarks

92 AI benchmarks with their saturation status, contamination risk and the source of every score. Filter the list, or scan the trust matrix below to see which capabilities still have benchmarks that separate frontier models.

Capability Domains
Status
Contamination
Sort By
 
Displaying 24 of 43 benchmarks (filtered from 92)
MathematicsActive

MathArena

A continuously refreshed platform evaluating models on newly released math competitions, proofs, research problems, and formalization.

Trust scoreout of 100
Low contamination
CodingActive

FrontierCode

Private maintainer-authored repository tasks graded for mergeability: correctness, tests, scope discipline, style, and code quality.

Trust scoreout of 100
Low contamination
MathematicsActive

Soohak

Fresh research-level mathematics authored by mathematicians, plus unanswerable prompts that reward justified refusal.

Trust scoreout of 100
Low contamination
MathematicsActive

FrontierMath

Original, unpublished research-level math problems by expert mathematicians — mostly private to resist contamination. Frontier models solved under 2% at launch.

Trust scoreout of 100
Low contamination
Instruction FollowingActive

FireBench

Evaluates instruction following for enterprise/API pipelines — exact output format, strict step ordering, ranking, and calibrated refusal across six categories.

Trust scoreout of 100
Low contamination
ReasoningActive

LiveBench (Continuous)

Monthly refreshed benchmark designed to prevent test-set contamination across reasoning, coding, mathematics, data analysis, and instruction following.

Trust scoreout of 100
Low contamination
Agentic & Tool UseActive

ARC-AGI-3

Interactive ARC environments that score agents by human-relative action efficiency, planning, exploration, and adaptation.

Trust scoreout of 100
Medium contamination
Instruction FollowingActive

IFBench

Verifiable instruction-following benchmark built from out-of-distribution constraints, designed to test generalisation beyond IFEval-style prompts.

Trust scoreout of 100
Medium contamination
Agentic & Tool UseActive

τ²-Bench Telecom

Customer-service agent benchmark in which both the agent and a simulated user act on a shared environment; this entry is the telecom domain.

Trust scoreout of 100
Medium contamination
Safety & AlignmentActive

HealthBench

Rubric-graded benchmark of realistic health conversations written with physicians, scoring the quality and safety of model responses.

Trust scoreout of 100
Medium contamination
KnowledgeActive

Humanity's Last Exam

~2,500 expert-written questions across 100+ subjects, each filtered to stump frontier models — designed to be the final closed-ended academic benchmark.

Trust scoreout of 100
Medium contamination
Multimodal VisionActive

MMMU-Pro

Harder, filtered variant of MMMU for multimodal college-level reasoning, with more answer options and questions embedded in images.

Trust scoreout of 100
Medium contamination
CodingActive

SciCode

Research-level scientific coding benchmark: models write code to solve subproblems drawn from real scientific workflows across natural-science domains.

Trust scoreout of 100
Medium contamination
Multimodal VisionActive

VIALS-AA

Agentic life-sciences benchmark requiring visual interpretation of scientific images.

Trust scoreout of 100
Medium contamination
Agentic & Tool UseActive

ITBench-AA

IT-operations benchmark covering site-reliability tasks such as root-cause analysis on Kubernetes systems.

Trust scoreout of 100
Medium contamination
Agentic & Tool UseActive

EnterpriseOps-Gym-AA

Enterprise tool-use benchmark of agents operating inside enterprise IT-service style environments.

Trust scoreout of 100
Medium contamination
Agentic & Tool UseActive

Harvey LAB-AA

Legal agent benchmark (Harvey LAB) run by Artificial Analysis.

Trust scoreout of 100
Medium contamination
Agentic & Tool UseActive

AA-AnalystAgent

Business and data-analyst agent benchmark evaluated by Artificial Analysis.

Trust scoreout of 100
Medium contamination
Agentic & Tool UseActive

AutomationBench-AA

Agentic SaaS workflow benchmark: automating multi-step business workflows across software tools.

Trust scoreout of 100
Medium contamination
Agentic & Tool UseActive

Multimodal Knowledge Agent (MKA)

Agentic knowledge-work benchmark that mixes multimodal inputs with tool use.

Trust scoreout of 100
Medium contamination
Long ContextActive

GDP.pdf

Professional-work benchmark over long PDF documents, evaluated by Artificial Analysis.

Trust scoreout of 100
Medium contamination
CodingActive

FrontierCode-AA 2.0

Second version of Artificial Analysis's FrontierCode software-engineering evaluation (rubric-graded real-world engineering tasks).

Trust scoreout of 100
Medium contamination
Agentic & Tool UseActive

APEX-Agents-AA

Professional-services agent benchmark evaluated by Artificial Analysis.

Trust scoreout of 100
Medium contamination

Trust matrix

Every benchmark by capability and status. Click a benchmark for its evidence and scores.

Contamination Risk
Refresh Cadence
Showing 92 of 92 benchmarks
CapabilityActiveNearing SaturationSaturatedDeprecated
Knowledge
Reasoning
Mathematics
Coding
Instruction Following
Long Context
Safety & Alignment
Multimodal Vision
Agentic & Tool Use