eval.yaml

name: APEX-Agents
description: >
  APEX-Agents is a benchmark from Mercor for evaluating whether AI agents can
  execute long-horizon, cross-application professional services tasks. Tasks
  were created by investment banking analysts, management consultants, and
  corporate lawyers, and require agents to navigate realistic work environments
  with files and tools (e.g., docs, spreadsheets, PDFs, email, chat, calendar).
evaluation_framework: archipelago

tasks:
  - id: apex-agents
    config: default
    split: train