FireBench: Evaluating Instruction Following in Enterprise and API-Driven LLM Applications
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866912943365423104 |
|---|---|
| author | Zhang, Yunfan Bei, Yijie Ravi, Jetashree Garbacki, Pawel |
| author_facet | Zhang, Yunfan Bei, Yijie Ravi, Jetashree Garbacki, Pawel |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_04857 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | FireBench: Evaluating Instruction Following in Enterprise and API-Driven LLM Applications Zhang, Yunfan Bei, Yijie Ravi, Jetashree Garbacki, Pawel Computation and Language Software Engineering 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. |
| title | FireBench: Evaluating Instruction Following in Enterprise and API-Driven LLM Applications |
| topic | Computation and Language Software Engineering |
| url | https://arxiv.org/abs/2603.04857 |