What it measures

Dynamic benchmark bridging open-ended user queries and ground-truth multiple-choice/short-answer benchmarks (0.96 correlation with Chatbot Arena).

Reading the measurement

Study what information the model receives and which inference the answer requires. A question may mix recall, visual perception, deduction and search. Use task breakdowns to find the mechanism behind a headline gain.

Published evaluation

Original dataset mixture and hard subset are distinct evaluations. Mixed-in benchmark subscores are not presented as full original benchmark results.

Where this fits in an evaluation suite

Ask what the task actually requires: recall, deduction, search, or a mixture. Separate closed-book reasoning from retrieval and tool-assisted problem solving.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Original dataset mixture and hard subset are distinct evaluations. Mixed-in benchmark subscores are not presented as full original benchmark results.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

  1. MixEval: Deriving Wisdom of the Crowd from LLM Benchmark Mixtures — published result table ↗

    Jinjie Ni, Fuzhao Xue, Xiang Yue, Yuntian Deng, Mahir Shah, Kabir Jain et al. · 2024

    All 8 authors

    Jinjie Ni, Fuzhao Xue, Xiang Yue, Yuntian Deng, Mahir Shah, Kabir Jain, Graham Neubig, Yang You

Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.