FireBench: Evaluating Instruction Following in Enterprise and API-Driven LLM Applications
This benchmark addresses a critical gap in evaluating LLMs for enterprise users, where strict adherence to output formats and content constraints is essential for reliable workflows.
This paper introduces FireBench, a new benchmark designed to evaluate instruction following in large language models (LLMs) specifically for enterprise and API-driven applications. It comprises over 2,400 samples across six capability dimensions, revealing key findings on the instruction following behavior of 11 evaluated LLMs in enterprise scenarios.
Instruction following is critical for LLMs deployed in enterprise and API-driven settings, where strict adherence to output formats, content constraints, and procedural requirements is essential for enabling reliable LLM-assisted workflows. However, existing instruction following benchmarks predominantly evaluate natural language generation constraints that reflect the needs of chat assistants rather than enterprise users. To bridge this gap, we introduce FireBench, an LLM instruction following benchmark grounded in real-world enterprise and API usage patterns. FireBench evaluates six core capability dimensions across diverse applications including information extraction, customer support, and coding agents, comprising over 2,400 samples. We evaluate 11 LLMs and present key findings on their instruction following behavior in enterprise scenarios. We open-source FireBench at fire-bench.com to help users assess model suitability, support model developers in diagnosing performance, and invite community contributions.