aiqbench analysis
This benchmark study audits when commercial LLMs choose to search the web versus answer from memory, using 400 questions split between widely known facts and post-cutoff news. GPT-5 mini and Grok 4 Fast search on over 90% of outside-knowledge questions but almost never on well-known facts, while Mistral Small often abstains rather than searching. For readers building agentic systems, the key takeaway is that search triggering varies sharply by product and depends partly on explicit temporal cues, so vendor behaviour should be audited rather than assumed.
Confident Fools or Careful Searchers? A Behavioural Study of When LLMs Search Beyond Their Static Internal Memory
arXiv:2610.41124v1 Announce Type: new Abstract: Agentic LLM assistants must repeatedly decide whether to answer from memory or to call a tool such as web search. Unnecessary searches cost latency and money; missing ones yield confident but unsupported answers. Prior work mostly measures verbalised confidence or trains new adaptive-retrieval methods, and rarely audits the search decisions that deployed products actually make. We introduce a benchmark of 400 multiple-choice questions that combines widely known inside-knowledge (IK) questions, which should be answered directly, with outside-knowledge (OK) questions about facts from news published after the models' training cut-offs (to our knowledge), which should trigger search or, without tools, abstention. Each item offers an abstain option, and two action-level scores, NAS and KIS, reward searching on OK questions and restraint on IK questions. Across five commercial LLMs with and without their vendors' web search, GPT-5 mini and Grok 4 Fast search on over 90% of OK but only 1% of IK questions, whereas Mistral Small searches on just 20% of OK questions, often abstaining instead of using its tool; one product's search indicator proved unusable. Searching is partly triggered by explicit temporal cues: on uncued OK questions, the NAS of the two selective models falls from about 0.72 to 0.57--0.58. The scores thus measure the search triggering of specific products, not metacognition in a strict sense.
Title:Confident Fools or Careful Searchers? A Behavioural Study of When LLMs Search Beyond Their Static Internal Memory
Abstract:Agentic LLM assistants must repeatedly decide whether to answer from memory or to call a tool such as web search. Unnecessary searches cost latency and money; missing ones yield confident but unsupported answers. Prior work mostly measures verbalised confidence or trains new adaptive-retrieval methods, and rarely audits the search decisions that deployed products actually make. We introduce a benchmark of 400 multiple-choice questions that combines widely known inside-knowledge (IK) questions, which should be answered directly, with outside-knowledge (OK) questions about facts from news published after the models' training cut-offs (to our knowledge), which should trigger search or, without tools, abstention. Each item offers an abstain option, and two action-level scores, NAS and KIS, reward searching on OK questions and restraint on IK questions. Across five commercial LLMs with and without their vendors' web search, GPT-5 mini and Grok 4 Fast search on over 90% of OK but only 1% of IK questions, whereas Mistral Small searches on just 20% of OK questions, often abstaining instead of using its tool; one product's search indicator proved unusable. Searching is partly triggered by explicit temporal cues: on uncued OK questions, the NAS of the two selective models falls from about 0.72 to 0.57--0.58. The scores thus measure the search triggering of specific products, not metacognition in a strict sense.
Artificial Intelligence (cs.AI)
arXiv:2610.41124 [cs.AI]
(or arXiv:2610.41124v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2610.41124 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both 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.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 327 words
Related coverage
Three types of model organism
LessWrong · Jun 10, 2026
Why Are LLMs Confidently Wrong? Correcting Overconfident Errors via Causal Head Intervention
arXiv · Oct 27, 2026
Beyond Calibration: Temporal LLM Confidence Assessment Across Languages and Local Settings
arXiv · Oct 21, 2026
Scenario Generation for Risk-Aware Reinforcement Learning with Probably Approximately Safe Guarantees
arXiv · Jun 8, 2026