Metodología de los Índices de Capacidad de Artificial Analysis
Resumen
Cada Índice de Capacidad de Artificial Analysis combina un conjunto seleccionado de evaluaciones en una única puntuación compuesta que mide el rendimiento de un modelo en un caso de uso profesional específico, como el trabajo legal, sanitario o financiero.
Ejecutamos de forma independiente cada evaluación que compone un índice antes de combinar sus puntuaciones en el índice. Como toda métrica de evaluación, los índices de capacidad tienen limitaciones y pueden no aplicarse directamente a cada caso de uso, pero ofrecen una síntesis útil para comparar modelos en el trabajo que importa a un dominio determinado.
Las evaluaciones subyacentes, incluyendo cómo se ejecuta y puntúa cada una, se documentan en la metodología de Benchmarking de Inteligencia.
Desglose de índices
Cada índice se centra en una sola profesión o campo, como Legal, Salud y Medicina, o Finanzas y Contabilidad, y pondera un conjunto de capacidades según la frecuencia con que cada una aparece en tareas reales de ese campo.
Nuestras ponderaciones de tareas se basan en una taxonomía de actividades laborales al estilo de O*NET. La tabla siguiente muestra los componentes y las ponderaciones de cada índice.
| Índice | Ponderación | Capacidad | Evaluaciones | Descripción |
|---|---|---|---|---|
| Finance & Accounting Index | 30% | Business Knowledge | AA-Omniscience | Domain recall in accounting, corporate finance, economics, and investments |
| 30% | Agentic Knowledge Work | GDPval-AA v2.1, AA-Briefcase v1.1 | Tool use, planning, and multi-step task execution, such as building spreadsheets, running an analysis, or coordinating a workflow | |
| 20% | Reasoning | HLE | Multi-step quantitative and analytic reasoning, used for sensitivity analysis, valuation, and structured problem solving | |
| 10% | Agentic Tool Use | AutomationBench-AA | Completing finance workflows across business apps such as spreadsheets, email, and accounting tools without breaking guardrails | |
| 5% | Long-Context | LCR, GDP.pdf | Reading and reasoning across long financial filings, deal documents, and PDF reports | |
| 5% | Non-Hallucination | AA-Omniscience | Avoiding fabricated figures or citations | |
| Strategy & Ops Index | 30% | Business Knowledge | AA-Omniscience | Working knowledge of business processes, accounting basics, and operational concepts |
| 35% | Agentic Knowledge Work | GDPval-AA v2.1, AA-Briefcase v1.1 | Tool use, planning, and orchestrating multi-step office workflows end-to-end | |
| 30% | Agentic Tool Use | AutomationBench-AA | Completing operations, HR, marketing, sales, and support workflows across business apps without breaking guardrails | |
| 5% | Long-Context | LCR, GDP.pdf | Holding context across long threads, policies, records, and PDF documents | |
| Legal Index | 35% | Legal Knowledge | AA-Omniscience | Recall of statutes, doctrines, and procedure across jurisdictions |
| 25% | Agentic Knowledge Work | GDPval-AA v2.1, AA-Briefcase v1.1 | Running matter-management workflows, drafting pipelines, and tool-augmented research | |
| 15% | Reasoning | HLE | Multi-step argumentation, statutory interpretation, and weighing conflicting authorities | |
| 10% | Long-Context | LCR, GDP.pdf | Reading and synthesizing across contracts, discovery productions, case-law packets, and PDF filings | |
| 10% | Non-Hallucination | AA-Omniscience | Avoiding fabricated case cites or invented statutes | |
| 5% | Agentic Tool Use | AutomationBench-AA | Completing client support and operations workflows across business apps without breaking guardrails | |
| Healthcare & Medical Index | 30% | Medical & Health Knowledge | AA-Omniscience | Clinical knowledge across diagnosis, pharmacology, and care pathways |
| 25% | Agentic Knowledge Work | GDPval-AA v2.1, AA-Briefcase v1.1 | Tool use, planning, and orchestrating EHR/pharmacy workflows end-to-end | |
| 15% | Long-Context Reasoning | MLCR-AA | Synthesising findings across long, fragmented patient records and claims files | |
| 10% | Non-Hallucination | AA-Omniscience | Avoiding fabricated drug interactions, doses, or guidelines | |
| 10% | Reasoning | HLE | Multi-step clinical reasoning across biology and medicine | |
| 10% | Agentic Tool Use | AutomationBench-AA | Completing patient support and operations workflows across business apps without breaking guardrails | |
| Engineering Index | 35% | Engineering Knowledge | AA-Omniscience | Domain recall across civil, electrical, mechanical, and other engineering disciplines |
| 30% | Reasoning | HLE, CritPt | Multi-step quantitative reasoning for derivations, sizing calculations, and design trade-offs | |
| 20% | Agentic Knowledge Work | GDPval-AA v2.1, AA-Briefcase v1.1 | Tool use, planning, and multi-step execution of engineering deliverables | |
| 15% | Agentic Terminal Use | Terminal-Bench 4.0 | Operating real terminal environments for builds, scripts, system administration, and debugging | |
| Economics Index | 35% | Economics Knowledge | AA-Omniscience | Recall across micro and macroeconomics, public finance, and markets |
| 35% | Reasoning | HLE | Multi-step quantitative and analytic reasoning for modeling, estimation, and inference | |
| 25% | Agentic Knowledge Work | GDPval-AA v2.1, AA-Briefcase v1.1 | Tool use, planning, and multi-step execution of analytical deliverables | |
| 5% | Long-Context Reasoning | LCR | Reading and reasoning across long reports, datasets, and research notes |