Research

Paper

AI LLM February 25, 2026

FewMMBench: A Benchmark for Multimodal Few-Shot Learning

Authors

Mustafa Dogan, Ilker Kesen, Iacer Calixto, Aykut Erdem, Erkut Erdem

Abstract

As multimodal large language models (MLLMs) advance in handling interleaved image-text data, assessing their few-shot learning capabilities remains an open challenge. In this paper, we introduce FewMMBench, a comprehensive benchmark designed to evaluate MLLMs under few-shot conditions, with a focus on In-Context Learning (ICL) and Chain-of-Thought (CoT) prompting. Covering a diverse suite of multimodal understanding tasks, from attribute recognition to temporal reasoning, FewMMBench enables systematic analysis across task types, model families, and prompting strategies. We evaluate 26 open-weight MLLMs from six model families across zero-shot, few-shot, and CoT-augmented few-shot settings. Our findings reveal that instruction-tuned models exhibit strong zero-shot performance but benefit minimally, or even regress, with additional demonstrations or CoT reasoning. Retrieval-based demonstrations and increased context size also yield limited gains. These results highlight FewMMBench as a rigorous testbed for diagnosing and advancing few-shot capabilities in multimodal LLMs. The data is available at: https://huggingface.co/datasets/mustafaa/FewMMBench

Metadata

arXiv ID: 2602.21854
Provider: ARXIV
Primary Category: cs.CL
Published: 2026-02-25
Fetched: 2026-02-26 05:00

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.21854v1</id>\n    <title>FewMMBench: A Benchmark for Multimodal Few-Shot Learning</title>\n    <updated>2026-02-25T12:30:18Z</updated>\n    <link href='https://arxiv.org/abs/2602.21854v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.21854v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>As multimodal large language models (MLLMs) advance in handling interleaved image-text data, assessing their few-shot learning capabilities remains an open challenge. In this paper, we introduce FewMMBench, a comprehensive benchmark designed to evaluate MLLMs under few-shot conditions, with a focus on In-Context Learning (ICL) and Chain-of-Thought (CoT) prompting. Covering a diverse suite of multimodal understanding tasks, from attribute recognition to temporal reasoning, FewMMBench enables systematic analysis across task types, model families, and prompting strategies. We evaluate 26 open-weight MLLMs from six model families across zero-shot, few-shot, and CoT-augmented few-shot settings. Our findings reveal that instruction-tuned models exhibit strong zero-shot performance but benefit minimally, or even regress, with additional demonstrations or CoT reasoning. Retrieval-based demonstrations and increased context size also yield limited gains. These results highlight FewMMBench as a rigorous testbed for diagnosing and advancing few-shot capabilities in multimodal LLMs. The data is available at: https://huggingface.co/datasets/mustafaa/FewMMBench</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-02-25T12:30:18Z</published>\n    <arxiv:comment>Preprint. 49 pages, 38 Figures, 5 Tables</arxiv:comment>\n    <arxiv:primary_category term='cs.CL'/>\n    <author>\n      <name>Mustafa Dogan</name>\n    </author>\n    <author>\n      <name>Ilker Kesen</name>\n    </author>\n    <author>\n      <name>Iacer Calixto</name>\n    </author>\n    <author>\n      <name>Aykut Erdem</name>\n    </author>\n    <author>\n      <name>Erkut Erdem</name>\n    </author>\n  </entry>"
}