Measuring Five-Nines Reliability: Sample-Efficient LLM Evaluation in Saturated Benchmarks

2026-05-11

AI Summary

Existing benchmarks struggle to efficiently evaluate the extreme reliability (e.g., 99.999%) required for real-world deployment of large language models, as conventional Monte Carlo methods incur prohibitive computational costs when estimating extremely low failure rates. This work introduces extreme reliability as a distinct evaluation dimension and proposes a sample-efficient assessment framework based on the Cross-Entropy Method (CEM), which learns an input sampling distribution focused on error-prone cases. Validated on GSM8K templates with models including Qwen2.5-Math-7B-Instruct, gpt-oss-20b-low, and Gemini 2.5 Flash Lite, the approach reduces inference queries by up to 156× compared to uniform sampling and uncovers significant reliability differences among models that standard benchmarks fail to distinguish, thereby demonstrating the necessity and feasibility of this evaluation paradigm.

Abstract

While existing benchmarks demonstrate the near-perfect performance of large language models (LLMs) on various tasks, this apparent saturation often obscures the need for rigorous evaluation of their reliability. In real-world deployment, however, achieving extremely high reliability (e.g., "five-nines" (99.999%) vs. "three-nines" (99.9%)) is fundamentally critical, as this gap results in an order-of-magnitude increase in failures, which is catastrophic in reliability-critical applications. Still, estimating such a rare failure probability with tight confidence bounds requires prohibitively large LLM inference sizes, making standard Monte Carlo evaluation infeasible under limited compute budgets. In this paper, we observe that LLM failures exhibit strong systematic patterns: across broad parameterized input spaces, a small subset of inputs disproportionately accounts for the majority of failures. Leveraging this observation, we propose to learn a sampling distribution concentrated on failure-prone inputs via the cross-entropy method (CEM). We evaluate our framework on three LLMs, Qwen2.5-Math-7B-Instruct, gpt-oss-20b-low, and Gemini 2.5 Flash Lite, across parameterized GSM8K templates and achieve up to 156.22x reduction in required inferences compared to naive uniform sampling. Our estimates reveal that models with indistinguishable accuracy on standard benchmarks can differ substantially in estimated failure rates, underscoring that reliability is a distinct and measurable axis of model quality. Our simple yet practical framework enables the evaluation of extreme reliability in LLMs, a distinct and underexplored dimension of evaluation beyond existing benchmarks, for their growing use in reliability-sensitive applications.

Problem

• LLM reliability
• five-nines reliability
• rare failure estimation
• saturated benchmarks
• sample-efficient evaluation

Innovation

• five-nines reliability
• sample-efficient evaluation
• cross-entropy method
• failure-prone inputs
• LLM reliability