MathArena
A continuously refreshed platform evaluating models on newly released math competitions, proofs, research problems, and formalization.
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.
A continuously refreshed platform evaluating models on newly released math competitions, proofs, research problems, and formalization.
Private maintainer-authored repository tasks graded for mergeability: correctness, tests, scope discipline, style, and code quality.
Fresh research-level mathematics authored by mathematicians, plus unanswerable prompts that reward justified refusal.
Original, unpublished research-level math problems by expert mathematicians — mostly private to resist contamination. Frontier models solved under 2% at launch.
Evaluates instruction following for enterprise/API pipelines — exact output format, strict step ordering, ranking, and calibrated refusal across six categories.
Monthly refreshed benchmark designed to prevent test-set contamination across reasoning, coding, mathematics, data analysis, and instruction following.
Interactive ARC environments that score agents by human-relative action efficiency, planning, exploration, and adaptation.
Verifiable instruction-following benchmark built from out-of-distribution constraints, designed to test generalisation beyond IFEval-style prompts.
Customer-service agent benchmark in which both the agent and a simulated user act on a shared environment; this entry is the telecom domain.
Rubric-graded benchmark of realistic health conversations written with physicians, scoring the quality and safety of model responses.
~2,500 expert-written questions across 100+ subjects, each filtered to stump frontier models — designed to be the final closed-ended academic benchmark.
Harder, filtered variant of MMMU for multimodal college-level reasoning, with more answer options and questions embedded in images.
Research-level scientific coding benchmark: models write code to solve subproblems drawn from real scientific workflows across natural-science domains.
Agentic life-sciences benchmark requiring visual interpretation of scientific images.
IT-operations benchmark covering site-reliability tasks such as root-cause analysis on Kubernetes systems.
Enterprise tool-use benchmark of agents operating inside enterprise IT-service style environments.
Legal agent benchmark (Harvey LAB) run by Artificial Analysis.
Business and data-analyst agent benchmark evaluated by Artificial Analysis.
Agentic SaaS workflow benchmark: automating multi-step business workflows across software tools.
Agentic knowledge-work benchmark that mixes multimodal inputs with tool use.
Medical long-context reasoning benchmark built on healthcare-claims style documents.
Professional-work benchmark over long PDF documents, evaluated by Artificial Analysis.
Second version of Artificial Analysis's FrontierCode software-engineering evaluation (rubric-graded real-world engineering tasks).
Professional-services agent benchmark evaluated by Artificial Analysis.
Every benchmark by capability and status. Click a benchmark for its evidence and scores.