RESEARCH · RESEARCH · #2310
EGI-BENCH: Salesforce AI Research publishes unified enterprise agent benchmark and EGI Score (arXiv v1)
Salesforce AI Research released EGI-BENCH, a unified command-line benchmark suite that consolidates eight enterprise-facing agent evaluations and introduces the six-dimensional Enterprise General Intelligence (EGI) Score with quick and deep runtime-budget modes. The paper reports quick-mode results for 14 LLMs (GPT, Claude, Gemini families), with top-line EGI ranging from 39.6 (GPT-4o mini) to 65.6 (gpt-5.5); code is available on GitHub.
KEY POINTS
- Salesforce AI Research released EGI-BENCH, a unified command-line benchmark suite that consolidates eight enterprise-facing agent evaluations and introduces the six-dimensional Enterprise General Intelligence (EGI) Score with quick and deep runtime-budget modes.
- The paper reports quick-mode results for 14 LLMs (GPT, Claude, Gemini families), with top-line EGI ranging from 39.6 (GPT-4o mini) to 65.6 (gpt-5.5); code is available on GitHub.
- A single, transparent enterprise-focused benchmark and multidimensional EGI Score make cross-model comparisons more consistent and help enterprises choose and stress-test LLMs for agentic workflows.
WHY IT MATTERS
A single, transparent enterprise-focused benchmark and multidimensional EGI Score make cross-model comparisons more consistent and help enterprises choose and stress-test LLMs for agentic workflows.