Research map / Large language models / Agents, code and math
LiveClin: A Live Clinical Benchmark without Leakage
Xidong Wang, Guo shuqi, Yue Shen, Junying Chen, Jian Wang, Jinjie Gu, Ping Zhang, Lei Liu, Benyou Wang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Critic agent to Generator (ablation) | Generator · Exam generation (200 case reports) | physician-validated accuracy (%): 84.5% -> 93.0% | Improved |
| add Critic agent to Generator (ablation) | Generator · Exam generation (200 case reports) | trivial question ratio (%): 16.5% -> 5.5% | Improved |
| replace physician authoring with Generator agent (ablation) | physician-authored questions · Exam generation (200 case reports) | trivial question ratio (%): 38.5% -> 16.5% | Improved |
| add Judge agent pre-screening to Generator-Critic (ablation) | Generator-Critic · Exam generation (200 case reports) | physician-validated accuracy (%): 93.0 -> 89.5 | Hurt |
| add Judge agent pre-screening to Generator-Critic (ablation) | Generator-Critic · Exam generation (200 case reports) | accuracy on Judge-approved subset (%): 89.5 -> 98.4 | Improved |
| use Generator agent instead of physician authoring (ablation) | physician authoring · Exam generation (200 case reports) | cost (USD per 100 cases): 4534.30 -> 35.34 | Improved |
| evaluate on cases published after model knowledge cutoff | GPT-5 · Pilot study case accuracy | Case Accuracy (%): 45.0% -> nearly 10 points lower | Hurt |
| newer release Claude 3.7 Sonnet instead of Claude 3.5 Sonnet | Claude 3.5 Sonnet · LiveClin | Case Accuracy (%) | Hurt |
| newer release Gemini 2.5 Flash instead of Gemini 2.0 Flash | Gemini 2.0 Flash · LiveClin | Case Accuracy (%) | Hurt |
| evaluate on August 2025 set instead of H1 2025 set | GPT-5 · LiveClin private leaderboard | Case Accuracy (%): 35.5 -> 34.1 | No clear effect |
Closest papers by meaning
- MedJourney: Benchmark and Evaluation of Large Language Models over Patient Clinical Journey (NeurIPS 2024 · similarity 0.86)
- PhysicianBench: Evaluating LLM Agents in Real-World EHR Environments (NeurIPS 2026 · similarity 0.85)
- LiveBench: A Challenging, Contamination-Limited LLM Benchmark (ICLR 2025 · similarity 0.85)
- MedCalc-Bench: Evaluating Large Language Models for Medical Calculations (NeurIPS 2024 · similarity 0.85)
- Position: Medical Large Language Model Benchmarks Should Prioritize Construct Validity (ICML 2025 · similarity 0.85)
- MTBBench: A Multimodal Sequential Clinical Decision-Making Benchmark in Oncology (NeurIPS 2025 · similarity 0.85)
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.
