{"slug": "the-benchmark-illusion-pruned-llms-can-pass-multiple-choice-but-fail-to-answer", "title": "The Benchmark Illusion: Pruned LLMs Can Pass Multiple Choice but Fail to Answer", "summary": "Researchers found that pruned large language models can pass multiple-choice benchmarks but fail to answer the same questions in open generation, creating a 'benchmark illusion.' The study, using multilingual question answering and high-sparsity pruning methods like Wanda, showed that correct answers are often demoted rather than erased, reappearing with beam search or sampling. This suggests compressed models should be tested on production tasks, not just recognition, to avoid overstating their usability.", "body_md": "# Computer Science > Computation and Language\n\n[Submitted on 16 Jun 2026]\n\n# Title:The Benchmark Illusion: Pruned LLMs Can Pass Multiple Choice but Fail to Answer\n\n[View PDF](/pdf/2606.17609)\n\n[HTML (experimental)](https://arxiv.org/html/2606.17609v1)\n\nAbstract:Compressing large language models reduces memory use and inference cost, but it can also create failures that standard benchmarks miss. A pruned model may still perform well on multiple-choice evaluations, yet fail to answer the same question in open generation. We ask what pruning changes: does it erase the correct answer, or does it make the answer harder to produce as the top output?\n\nWe study this question with multilingual question answering, tracking the same questions before and after pruning. We find a benchmark illusion. Under high-sparsity pruning, especially Wanda, models often fail in greedy open generation while still selecting the correct answer under multiple-choice scoring. In these recognition-only errors, the answer is usually not gone, but demoted: it often reappears with beam search, sampling, or one in-context example. Overall, multiple-choice benchmarks can overstate the usability of compressed LLMs, creating an evaluation blind spot. Compressed models should be tested on what they can produce, not only on what they can recognize.\n\n### References & Citations\n\nLoading...\n\n# Bibliographic and Citation Tools\n\nBibliographic Explorer\n\n*(*[What is the Explorer?](https://info.arxiv.org/labs/showcase.html#arxiv-bibliographic-explorer))\nConnected Papers\n\n*(*[What is Connected Papers?](https://www.connectedpapers.com/about))\nLitmaps\n\n*(*[What is Litmaps?](https://www.litmaps.co/))\nscite Smart Citations\n\n*(*[What are Smart Citations?](https://www.scite.ai/))# Code, Data and Media Associated with this Article\n\nalphaXiv\n\n*(*[What is alphaXiv?](https://alphaxiv.org/))\nCatalyzeX Code Finder for Papers\n\n*(*[What is CatalyzeX?](https://www.catalyzex.com))\nDagsHub\n\n*(*[What is DagsHub?](https://dagshub.com/))\nGotit.pub\n\n*(*[What is GotitPub?](http://gotit.pub/faq))\nHugging Face\n\n*(*[What is Huggingface?](https://huggingface.co/huggingface))\nScienceCast\n\n*(*[What is ScienceCast?](https://sciencecast.org/welcome))# Demos\n\n# Recommenders and Search Tools\n\nInfluence Flower\n\n*(*[What are Influence Flowers?](https://influencemap.cmlab.dev/))\nCORE Recommender\n\n*(*[What is CORE?](https://core.ac.uk/services/recommender))# arXivLabs: experimental projects with community collaborators\n\narXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.\n\nBoth individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.\n\nHave an idea for a project that will add value for arXiv's community? [ Learn more about arXivLabs](https://info.arxiv.org/labs/index.html).", "url": "https://wpnews.pro/news/the-benchmark-illusion-pruned-llms-can-pass-multiple-choice-but-fail-to-answer", "canonical_source": "https://arxiv.org/abs/2606.17609", "published_at": "2026-06-17 01:53:22+00:00", "updated_at": "2026-06-17 02:23:14.640486+00:00", "lang": "en", "topics": ["large-language-models", "ai-safety", "ai-research", "natural-language-processing"], "entities": ["Wanda"], "alternates": {"html": "https://wpnews.pro/news/the-benchmark-illusion-pruned-llms-can-pass-multiple-choice-but-fail-to-answer", "markdown": "https://wpnews.pro/news/the-benchmark-illusion-pruned-llms-can-pass-multiple-choice-but-fail-to-answer.md", "text": "https://wpnews.pro/news/the-benchmark-illusion-pruned-llms-can-pass-multiple-choice-but-fail-to-answer.txt", "jsonld": "https://wpnews.pro/news/the-benchmark-illusion-pruned-llms-can-pass-multiple-choice-but-fail-to-answer.jsonld"}}