When Calibration Rankings Reverse: Accuracy-Controlled Evaluation for Fair Comparison of LLMs

arXiv cs.CL Papers

Summary

This paper demonstrates that global calibration metrics like Expected Calibration Error are confounded by model accuracy, and proposes ACE, an accuracy-controlled evaluation framework for fair comparison of large language models.

arXiv:2606.30814v1 Announce Type: new Abstract: Calibration evaluates whether a model confidence aligns with its empirical accuracy. Existing studies often compare the calibration of different large language models using global calibration metrics such as Expected Calibration Error and Brier Score. We begin by showing, both theoretically and empirically, that such comparisons are confounded by differences in model accuracy. For fairer cross-model comparison, we then propose ACE, an accuracy-controlled evaluation framework with three complementary views: Instance-Aligned, Distribution-Aligned, and Candidate-Aligned calibration. Across multiple benchmarks, model families, and confidence elicitation methods, we use ACE to study two practically important comparison axes, small versus large models and thinking versus non-thinking models. We find that many previously reported calibration advantages under raw global metrics weaken substantially after accuracy control. We also find that ranking reversal is frequent: models favored by raw metrics often cease to be favored once accuracy is controlled. Our results show that raw global calibration metrics are not robust for cross-model comparison, and that fair calibration comparison requires accuracy-aware evaluation.
Original Article
View Cached Full Text

Cached at: 07/01/26, 05:31 AM

# When Calibration Rankings Reverse: Accuracy-Controlled Evaluation for Fair Comparison of LLMs
Source: [https://arxiv.org/abs/2606.30814](https://arxiv.org/abs/2606.30814)
[View PDF](https://arxiv.org/pdf/2606.30814)

> Abstract:Calibration evaluates whether a model confidence aligns with its empirical accuracy\. Existing studies often compare the calibration of different large language models using global calibration metrics such as Expected Calibration Error and Brier Score\. We begin by showing, both theoretically and empirically, that such comparisons are confounded by differences in model accuracy\. For fairer cross\-model comparison, we then propose ACE, an accuracy\-controlled evaluation framework with three complementary views: Instance\-Aligned, Distribution\-Aligned, and Candidate\-Aligned calibration\. Across multiple benchmarks, model families, and confidence elicitation methods, we use ACE to study two practically important comparison axes, small versus large models and thinking versus non\-thinking models\. We find that many previously reported calibration advantages under raw global metrics weaken substantially after accuracy control\. We also find that ranking reversal is frequent: models favored by raw metrics often cease to be favored once accuracy is controlled\. Our results show that raw global calibration metrics are not robust for cross\-model comparison, and that fair calibration comparison requires accuracy\-aware evaluation\.

## Submission history

From: Ruihan Yang \[[view email](https://arxiv.org/show-email/6c7a7840/2606.30814)\] **\[v1\]**Mon, 29 Jun 2026 18:37:48 UTC \(9,358 KB\)

Similar Articles

Confidence Calibration in Large Language Models

arXiv cs.AI

This paper analyzes the confidence calibration of 11 popular LLMs, finding that they are generally overconfident, especially on hard tasks, and underconfident on easy tasks. It introduces LifeEval, a test for evaluating calibration across difficulty levels.