AI Paper Atlas

Research map / Large language models / Agents, code and math

Benchmarking LLM Tool-Use in the Wild

Peijie Yu, Wei Liu, Yifan Yang, Jinjian Li, Zelong Zhang, Xiao Feng, feng zhang

ICLR 2026 · Large language models › Agents, code and math · approach: Software engineering agents · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
move from task order 1 to task order 4 within a dialogueGemini-2.0-Thinking · WildToolBenchtask accuracy (%): 78.13 -> 50.39Hurt
move from task order 1 to task order 4 within a dialogueQwen2.5-0.5B-Instruct · WildToolBenchtask accuracy (%): 43.75 -> 25.39Hurt
scale Qwen2.5-Instruct from 0.5B to 72BQwen2.5-Instruct · WildToolBenchsession accuracy (%): 0.78 -> 6.25Improved
scale Qwen2.5-Instruct from 3B to 7BQwen2.5-Instruct · WildToolBenchsession accuracy (%): 4.30 -> 4.30No clear effect
enable thinking modeQwen3-14B · WildToolBenchsession accuracy (%): 7.81 -> 9.38Improved
enable thinking modeQwen3-32B · WildToolBenchsession accuracy (%): 8.59 -> 7.81Hurt
enable thinking modeDoubao-1.5 · WildToolBenchsession accuracy (%): 0.78 -> 6.64Improved
enable thinking modeDoubao-1.6 · WildToolBenchsession accuracy (%): 7.03 -> 3.13Hurt
use tool-specialized model instead of general model (xLAM-2-70B vs GLM-4.5)tool-use LLMs · WildToolBenchoptimal path rate, overall (%): 39.42 -> 19.92Hurt
tasks requiring clarification instead of casual chatGrok-4 · WildToolBenchtask accuracy (%): 66.02 -> 33.59Hurt
mixed sequential+parallel instead of parallel-only compositional tasksClaude-4-Sonnet · WildToolBenchtask accuracy (%): 54.78 -> 25.00Hurt
more instruction transitions in dialogueLLMs overall · WildToolBenchtask accuracy: drop as much as 30%Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy