AutoBench: Automating LLM Evaluation through Reciprocal Peer Assessment

TL;DR

AutoBench introduces an automated, peer-assessment framework for evaluating large language models that dynamically generates tasks and aggregates judgments, providing a scalable and contamination-resistant alternative to static benchmarks.

Contribution

It presents a novel reciprocal peer assessment methodology for LLM evaluation, enabling dynamic task generation and consensus-based ranking, validated through strong correlation with established benchmarks.

Findings

AutoBench correlates 78% with MMLU-Pro and 63% with GPQA.

Multi-judge evaluations outperform single-judge baselines.

Framework is scalable and resistant to test-set contamination.

Abstract

We present AutoBench, a fully automated and self-sustaining framework for evaluating Large Language Models (LLMs) through reciprocal peer assessment. This paper provides a rigorous scientific validation of the AutoBench methodology, originally developed as an open-source project by eZecute S.R.L.. Unlike static benchmarks that suffer from test-set contamination and limited adaptability, AutoBench dynamically generates novel evaluation tasks while models alternately serve as question generators, contestants, and judges across diverse domains. An iterative weighting mechanism amplifies the influence of consistently reliable evaluators, aggregating peer judgments into consensus-based rankings that reflect collective model agreement. Our experiments demonstrate strong correlations with established benchmarks including MMLU-Pro and GPQA (respectively 78\% and 63\%), validating this peer-driven evaluation paradigm. The multi-judge design significantly outperforms single-judge baselines, confirming that distributed evaluation produces more robust and human-consistent assessments. AutoBench offers a scalable, contamination-resistant alternative to static benchmarks for the continuous evaluation of evolving language models.