OpenAI Unveils GPT‑Rosalind: A New AI Model for Accelerating Life‑Science Research

Model Overview

On 16 April 2026 OpenAI released GPT‑Rosalind, a reasoning model for life‑science research (biology, drug discovery, translational medicine). The model is optimized for scientific workflows such as evidence synthesis, hypothesis generation, experimental design and multi‑step analysis.

Motivation

In the United States, bringing a new drug from target discovery to regulatory approval averages 10–15 years. Progress is limited by the complexity of research workflows that require navigating large literature corpora, specialized databases, experimental data and evolving hypotheses. Accelerating these workflows can reduce time and increase success rates.

Evaluation Framework

The team assessed GPT‑Rosalind across core scientific sub‑domains: chemical reaction mechanisms; protein structure, mutation effects and interactions; DNA phylogenetic interpretation. Additional tests measured the ability to interpret experimental results, recognize expert patterns, design follow‑up experiments, and select appropriate computational tools, databases and domain‑specific capabilities. The evaluation aimed to determine whether the model can support end‑to‑end research workflows.

Benchmark Results

On the BixBench benchmark (real‑world bioinformatics and data‑analysis tasks), GPT‑Rosalind achieved the highest scores among publicly released models.

On the LABBench2 suite (literature retrieval, database access, sequence manipulation, protocol design), GPT‑Rosalind outperformed GPT‑5.4 on 6 of 11 tasks. The largest gain was on the CloningQA task, which requires designing DNA and enzyme protocols for molecular cloning.

Industry Evaluation

In collaboration with Dyno Therapeutics, GPT‑Rosalind was evaluated on unpublished RNA sequences. Compared with scores from 57 AI‑biology experts, the model ranked in the 95th percentile for prediction tasks and around the 84th percentile for sequence‑generation tasks.

Tool Integration

A life‑science research plugin for Codex provides more than 50 public multi‑omics databases, literature sources and bio‑tools. The plugin supplies entry points for common reproducible tasks such as protein‑structure lookup, sequence search, literature review and public‑dataset discovery.

Access and Deployment

Research organizations can request access through a qualification and safety‑review process. During the preview period the model does not consume existing credit or token balances, though usage is subject to abuse‑prevention safeguards.

Plugin repository: https://github.com/openai/plugins/tree/main/plugins/life-science-research

Access form: https://openai.com/form/life-sciences-access/

Code example

来源：ScienceAI
本文
约2000字
，建议阅读
5
分钟
一个旨在加速科学研究和药物发现的全新模型系列。