Same Outputs, Different Scores: Auditing the Unchecked Graders of Multimodal LLM Benchmarks
Factual evidence
What the source reports
Researchers demonstrate that unexamined scoring code in multimodal LLM benchmarks distorts published performance scores.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.CL — Computation and Language · RESEARCH
- Published by source
- 4 November 2026
- Collected by OneBench
- 5 Nov 2026, 02:01 UK
Stored source excerpt
arXiv:2611.03377v1 Announce Type: cross Abstract: Every multimodal-LLM multiple-choice benchmark ends with a scoring rule that converts a model's response, often free-form text rather than a…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.
OneBench interpretation
Institutional assessment
So what
Published vendor benchmarks for multimodal vision-language models may be unreliable due to inconsistent, unstandardized internal grading code.
Do what
Review internal model evaluation pipelines to ensure model grading functions parse free-form outputs deterministically and consistently across benchmark tests.