# MedCalc-Bench Doesn't Measure What You Think: A Benchmark Audit and the Case for Open-Book Evaluation

arXiv: https://arxiv.org/abs/2603.02222 · HTML: https://mlsanity.com/paper/2603.02222/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Placed in Mathematical reasoning, one of 17 directions in LLM reasoning.
- Builds on 3 of its 8 references that are in the map.
- First posted on arXiv 10 Feb 2026.

## Placed in

- LLM reasoning / [Mathematical reasoning](https://mlsanity.com/lr/mathematical-reasoning/)

## Builds on (most cited first, 3 of 3)

- [MedCalc-Bench: Evaluating Large Language Models for Medical Calculations](https://mlsanity.com/paper/2406.12036/) (2024)
- [RiskAgent: Synergizing Language Models with Validated Tools for Evidence-Based Risk Prediction](https://mlsanity.com/paper/2503.03802/) (2025)
- [Scalable Stewardship of an LLM-Assisted Clinical Benchmark with Physician Oversight](https://mlsanity.com/paper/2512.19691/) (2025)

## Built on by (most cited first, 3 of 3)

- [Capabilities of Claude Fable 5 on Biomedical Challenge Problems](https://mlsanity.com/paper/2607.10849/) (2026)
- [LabFactory: Building and Evaluating Executable AI Labs](https://mlsanity.com/paper/2609.28697/) (2026)
- [Unknown is not normal: separating language-model extraction from rule-based decision logic for clinical risk scores](https://mlsanity.com/paper/2609.34112/) (2026)

## Abstract (arXiv)

MedCalc-Bench is a widely used benchmark for evaluating LLM performance on clinical calculator tasks, with state-of-the-art direct prompting scores plateauing around 35% on the Verified split (HELM MedHELM leaderboard) and the best published approach-RL with verifiable rewards-reaching 74%. We present three contributions that challenge the benchmark's current framing. First, we conduct a systematic audit of the benchmark's calculator implementations, identifying and fixing over 20 errors ranging from critical formula inaccuracies to runtime bugs in a NeurIPS-published dataset. Second, we show that a simple intervention-providing the model with the calculator specification at inference time ("open-book" prompting)-raises accuracy from ~52% to 81-85% on GLM-4.6V and GLM-4.7, surpassing all published results including RL-trained systems, without any fine-tuning. Third, we establish an upper bound of 95-97% using GPT-5.2-Thinking, with residual errors attributable primarily to ground-truth issues and dataset ambiguities. Our findings suggest that MedCalc-Bench predominantly measures formula memorization and arithmetic precision rather than clinical reasoning, and would be better framed as a tool-use evaluation.