# OpenAI Introduces MedChemBench to Test AI on Medicinal Chemistry Reasoning

> A new benchmark of 99 problems drawn from 117 papers and patents evaluates how well models connect molecular structure to experimental evidence.

- Publication: Thoai Dispatch
- Canonical URL: https://vokythoai.dev/article/openai-medchembench-medicinal-chemistry-benchmark
- Section: News
- Author: OpenAI Blog
- Published: 2026-11-20
- Tags: openai, medchembench, benchmark, medicinal-chemistry, drug-discovery, gpt-6, reasoning, molecular-structure, research

OpenAI has introduced MedChemBench, a benchmark that evaluates AI models on reasoning tasks drawn from medicinal chemistry research, according to [the company's announcement](https://openai.com/index/introducing-medchembench). The benchmark tests whether models can connect a molecule's structure with experimental evidence to explain results, compare compounds, and predict their properties. OpenAI frames the work as one step toward building AI that can reliably support scientists developing medicines.

## Why it matters

Medicinal chemistry involves tradeoffs that are difficult to reason about. Improving one property of a molecule can make another worse: a compound might act strongly against a disease-related protein but be poorly absorbed or break down too quickly in the body. Medicinal chemists work through these tradeoffs by designing, making, and testing related compounds, then deciding what to try next. For AI to help, it needs to connect changes in molecular structure with experimental evidence and reason about which compounds are worth pursuing. MedChemBench is designed to measure those abilities rather than general chemistry knowledge.

## Technical details

MedChemBench comprises 99 problems drawn from published research, built from 117 papers and patents and including 761 distinct molecular structures. For most problems, models receive the context available at a chosen point in a research program: the biological target, the researchers' goals, the compounds studied so far, and selected experimental findings.

The benchmark grades four components:

- **Molecular structures:** correct atom connections and specified stereochemistry in molecules assembled from provided drawings and chemical groups.
- **Scientific reasoning:** explanations supported by the supplied evidence, such as reasons for choosing one compound over another.
- **Ranking and selection:** compound rankings and choices that meet stated requirements.
- **Property predictions:** agreement with numerical values and categories reported in the source material, including measured and calculated molecular properties.

At the xhigh reasoning setting, GPT-6.2 Astra earned 84.5% of the available credit, compared with 83.3% for GPT-6 Astra, 82.0% for GPT-6.1 Sol, and 58.2% for GPT-5.6 Sol. Across the full benchmark at the shared xhigh setting, Astra earned 83.3% of the available credit, compared with 82.0% for GPT-6.1 Sol, 75.1% for GPT-6.1 Terra, and 58.2% for GPT-5.6 Sol. At max effort, GPT-6 Astra reached 83.8%. OpenAI evaluated eleven GPT models at different reasoning-effort settings, reporting 51 combinations across nine models from GPT-5.6 onward. Higher reasoning settings improved scores, while large differences between models remained.

On the structure component, Astra earned 95.4% of the available credit at xhigh. One problem asks models to assemble a molecule from a central structure and two attached groups while preserving an 18-atom ring. GPT-6 Astra reconstructed the compound correctly in all ten responses, while GPT-5.6 Sol and GPT-5.6 Luna failed across all ten. In the illustrated answers, GPT-5.6 Sol produced a 15-atom ring, and GPT-5.6 Luna split the molecule into two disconnected fragments. OpenAI notes that such errors change the identity of the molecule being analyzed.

A separate problem draws on research into inhibitors of PRMT5, a protein studied in cancer drug discovery. The model must predict how three related compounds will block PRMT5 in cells, cross a cell barrier, and be broken down by liver enzymes. Later experiments, withheld from the model, showed a tradeoff: compound 23 blocked PRMT5 at a lower concentration, while compound 24 crossed the cell barrier more readily and broke down more slowly. Predictions receive partial credit based on closeness to measured values, and measurements supplied in the prompt must be reproduced correctly.

## What to watch

OpenAI reports that GPT-6 Astra was strongest at reconstructing molecular structures and explaining experimental evidence, while selecting compounds and predicting numerical properties proved harder for that model. The component breakdown separates four abilities, showing where models diverge. The results suggest that structure reconstruction is close to saturated for the strongest models, while compound selection and quantitative property prediction remain the harder parts of the task. How those gaps close in future models will be the practical signal for whether AI can meaningfully assist medicinal chemists.

## Sources

- [OpenAI Blog](https://openai.com/index/introducing-medchembench) (2026-11-20)
