Benchmark RadarCEO-Bench
CEO-Bench evaluates long-horizon agent capabilities by simulating a startup over 500 days. Agents manage pricing, marketing, budgeting, and other business aspects through a programmable Python interface, facing noisy data and changing market conditions. Performance is measured by final company balance against a rule-based baseline.
- Released
- 2026-06-16
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
CEO-Bench addresses the evaluation gap for agents that must sustain adaptive progress over long horizons, combining uncertainty, information acquisition, and multi-step coordination. It provides a decision-useful benchmark for comparing models on realistic, dynamic business management tasks.
Motivation
Language model agents are becoming proficient executors at isolated, short-horizon tasks such as software engineering and customer service.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.