---
title: 'GPT-Rosalind: OpenAI''s Life Sciences AI Upgrade (2026)'
description: OpenAI's June 2026 GPT-Rosalind update beats GPT-5.5 on drug-discovery benchmarks while using fewer tokens, adds Codex plugins, and partners with Novo Nordisk.
url: https://nerdleveltech.com/gpt-rosalind-openai-life-sciences-update
language: en
date: '2026-06-05'
category: ai-ml
tags:
  - gpt-rosalind
  - openai
  - drug discovery
  - life sciences ai
  - medchembench
  - codex plugins
  - novo nordisk
markdown_url: https://nerdleveltech.com/gpt-rosalind-openai-life-sciences-update.md
attribution: 'Source: NerdLevelTech — https://nerdleveltech.com/gpt-rosalind-openai-life-sciences-update'
terms: https://nerdleveltech.com/llms.txt
---

# GPT-Rosalind: OpenAI's Life Sciences AI Upgrade (2026)

On June 3, 2026, OpenAI shipped a major update to **GPT-Rosalind**, its life-sciences reasoning model, folding in GPT-5.5's agentic coding and tool use alongside deeper intelligence in medicinal chemistry and genomics. On OpenAI's own benchmarks the updated model beats GPT-5.5 across medicinal chemistry, genomics, and wet-lab tasks — while spending fewer tokens to get there — and it's now in research preview for eligible organizations worldwide.[^1]

## TL;DR

GPT-Rosalind is OpenAI's first purpose-built life-sciences model, made for biology, drug discovery, and translational medicine. The June 3, 2026 update layers GPT-5.5's coding and tool-use abilities onto stronger domain reasoning, and OpenAI reports gains over GPT-5.5 on three new in-house benchmarks: MedChemBench (27.5% vs. 25.1%), GeneBench (21.6% vs. 20.4%), and LabWorkBench (63.2% vs. 55.8%), each with lower token usage.[^1] Two Codex plugins — Life Sciences Research and Life Sciences NGS Analysis — turn that reasoning into runnable bioinformatics workflows, and OpenAI spotlighted Novo Nordisk among its named life-sciences partners, a roster that also includes Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific.[^1][^2] The model stays in research preview under a trusted-access program, and the raw scores are a reminder that this is acceleration for expert scientists, not autonomous drug design.

## What is GPT-Rosalind?

GPT-Rosalind is a frontier reasoning model that OpenAI purpose-built for research across biology, drug discovery, and translational medicine. OpenAI first introduced it on April 16, 2026 as the first release in its life-sciences model series, deployed through a "trusted-access" structure rather than a general public launch.[^2] Where a general model like GPT-5.5 is optimized for broad tasks, GPT-Rosalind is tuned to reason over molecules, proteins, genes, pathways, and disease-relevant biology, and to drive the scientific tools and databases researchers actually use.[^1]

The model is named after Rosalind Franklin, whose research helped reveal the structure of DNA.[^2] The positioning is narrow on purpose: this is a model for people doing literature review, sequence-to-function interpretation, experimental planning, and data analysis — not a chatbot for everyone.[^1]

## What's new in the June 2026 GPT-Rosalind update

The headline change is the foundation. The updated GPT-Rosalind integrates the agentic coding and tool-use capabilities of **GPT-5.5**, the model OpenAI released on April 23, 2026.[^1][^3] On top of that, OpenAI added stronger intelligence in two core drug-discovery domains — medicinal chemistry and genomics — and broadened performance across analysis, design, and experimental workflows.[^1]

To measure that, OpenAI built **LifeSciBench**, an externally expert-judged benchmark that takes an end-to-end view of research rather than testing one capability in isolation. It draws tasks from six workflow areas central to life-sciences work: evidence handling, analysis, design and optimization, scientific reasoning, validation and operations, and translation and communication.[^1] The intent is to align model progress with the realities of how scientific work actually gets done, instead of optimizing for a single narrow score.

## GPT-Rosalind vs GPT-5.5: the benchmarks

OpenAI reports that the specialized model outperforms its own GPT-5.5 base across three new domain benchmarks — and, notably, does so while using fewer tokens, which matters for both cost and latency on long-horizon scientific tasks.[^1]

| Benchmark | What it measures | GPT-Rosalind | GPT-5.5 | Token efficiency |
|---|---|---|---|---|
| MedChemBench | Medicinal chemistry: SAR, ADME, lead optimization, retrosynthesis | 27.5% | 25.1% | 7.2% fewer tokens |
| GeneBench | Long-horizon genomics and quantitative biology analysis | 21.6% | 20.4% | 31% fewer tokens |
| LabWorkBench | Linking perturbations to outcomes in real wet-lab protocols | 63.2% | 55.8% | 5.3% fewer tokens |

The standout is LabWorkBench, where GPT-Rosalind scores 63.2% against GPT-5.5's 55.8% — a 7.4-point gain on a benchmark OpenAI says uses proprietary, uncontaminated wet-lab protocol data.[^1] The genomics result is more modest in accuracy (a 1.2-point gain) but pairs with the largest efficiency win: 31% fewer tokens than GPT-5.5 on GeneBench's long-horizon agentic tasks.[^1]

One important caveat on framing: every one of these is an OpenAI-designed benchmark, with scores measured by OpenAI (LifeSciBench is externally expert-judged, but the others are in-house evaluations). They are not yet independently reproduced, so they're best read as internal progress signals rather than settled, cross-lab rankings.

## From reasoning to executed workflows: the Codex plugins

A reasoning model that only talks isn't much use in a lab. With this update, OpenAI shipped two plugins that give GPT-Rosalind an execution layer inside **Codex**: **Life Sciences Research** and **Life Sciences NGS Analysis**.[^1] Together they bring sourced evidence retrieval, biological interpretation, and bioinformatics execution into one workspace, so researchers can connect external evidence with their own omics analyses while preserving artifacts and provenance.[^1]

Access is tiered. All Codex users can use both plugins; qualified GPT-Rosalind enterprise users can additionally use the GPT-Rosalind model itself to power them.[^1] OpenAI also added interactive viewers for biologically native file types — sequence, alignment, and structure — so a scientist can inspect a mutant residue or an inhibitor-bound pocket directly while the model reasons across the workflow.[^1] In OpenAI's demo, the NGS Analysis plugin turns a review of processed ctDNA records into an interactive notebook that surfaces a recurring KRAS G12C alteration, then the Research plugin layers in target, inhibitor, and resistance context.[^1]

## Who gets access — and the Novo Nordisk partnership

GPT-Rosalind remains in **research preview** under a trusted-access deployment model. Eligibility is restricted to organizations conducting legitimate research with clear public benefit, strong governance and safety oversight, and controlled, enterprise-grade access.[^1] With this update, OpenAI expanded availability to eligible organizations globally and began offering an OpenAI-managed workspace for qualified organizations that don't have an Enterprise account.[^1]

The partner OpenAI spotlighted in this update is **Novo Nordisk**, the maker behind blockbuster GLP-1 drugs. "To deliver meaningful value for researchers, advanced AI models must be grounded in trusted scientific data, connected to validated tools, and integrated into the real-world workflows researchers use every day," said Mishal Patel, Group Vice President of AI & Digital Innovation, R&D, at Novo Nordisk.[^1] Novo Nordisk was already among the customers OpenAI named at GPT-Rosalind's April launch — a roster that also includes Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific.[^2] OpenAI also tied the work to its broader public-benefit agenda, including Rosalind Biodefense, which it detailed on May 29, 2026.[^1]

## GPT-Rosalind vs Isomorphic Labs: two bets on AI drug discovery

OpenAI isn't alone in chasing AI for biology, and the contrast with Google's camp is instructive. **Isomorphic Labs**, the drug-discovery company spun out of Google DeepMind and built on AlphaFold, raised $2.1 billion in a Series B round in May 2026 to push its pipeline of AI-designed drugs toward the clinic.[^4]

The two represent different theories of where AI adds value. Isomorphic Labs is building an end-to-end drug-design engine aimed at producing actual clinical candidates. GPT-Rosalind is a reasoning layer that accelerates the human research process — synthesizing evidence, critiquing study designs, planning experiments, and running bioinformatics — rather than autonomously designing molecules.[^1][^4] One aims to replace parts of the discovery pipeline; the other aims to make the scientists running it faster.

## The reality check: what 27.5% actually means

It's worth sitting with that MedChemBench number. A 27.5% score means GPT-Rosalind still fails roughly three out of four medicinal-chemistry tasks on OpenAI's own benchmark.[^1] That's not a knock — these are deliberately hard, expert-designed evaluations, and beating a frontier general model by a few points while using fewer tokens is a real result. But it frames the honest use case: GPT-Rosalind is a tool that helps expert researchers move faster and pressure-test their own work, not a system that designs drugs on its own.

OpenAI's own framing supports this. The model is pitched as a "more capable partner across the full life cycle of scientific research," helping scientists move "from the right questions to clearer evidence, better experiments."[^1] The eval example OpenAI published — a hard-nosed critique of an FDA accelerated-approval package for a gene therapy — is telling: the model's value there is catching the assay-specificity flaws and confounders a skeptical reviewer would raise, not inventing the therapy.[^1]

## FAQ

**What is GPT-Rosalind?**
GPT-Rosalind is OpenAI's purpose-built reasoning model for life sciences — biology, drug discovery, and translational medicine. First launched April 16, 2026, it's tuned to reason over molecules, genes, proteins, and pathways and to operate scientific tools and databases.[^1][^2]

**What changed in the June 2026 GPT-Rosalind update?**
The update integrates GPT-5.5's agentic coding and tool use, adds stronger medicinal-chemistry and genomics reasoning, ships two Codex plugins (Life Sciences Research and NGS Analysis), expands access globally, and spotlights Novo Nordisk as a partner.[^1]

**How does GPT-Rosalind compare to GPT-5.5?**
On OpenAI's benchmarks, GPT-Rosalind beats GPT-5.5 on MedChemBench (27.5% vs. 25.1%), GeneBench (21.6% vs. 20.4%), and LabWorkBench (63.2% vs. 55.8%), each while using fewer tokens.[^1]

**Can anyone use GPT-Rosalind?**
No. The model stays in research preview under a trusted-access program limited to vetted organizations with legitimate research aims and strong governance. The two Codex plugins, however, are available to all Codex users.[^1]

**Is GPT-Rosalind the same as Isomorphic Labs?**
No. Isomorphic Labs (a Google DeepMind spinout) is building an end-to-end drug-design platform aimed at clinical candidates. GPT-Rosalind is a reasoning model that accelerates human research workflows rather than autonomously designing drugs.[^1][^4]

## Bottom line

GPT-Rosalind's June 2026 update is a measured, credible step: a GPT-5.5 foundation, real (if early) benchmark gains over that base, an execution layer in Codex, and a serious pharma partner in Novo Nordisk. The benchmark scores — beating GPT-5.5 while spending fewer tokens — are the kind of efficiency-plus-quality win that matters for long, tool-heavy scientific work. Just keep the framing honest: with a 27.5% MedChemBench pass rate, this is a model that makes expert scientists faster, not one that does the science for them.

---

*Related reading: [OpenAI's GPT-5.5-Cyber and the AISI frontier-parity evaluation](aisi-gpt-5-5-cyber-evaluation-frontier-parity), [building smarter apps with the OpenAI API](building-smarter-apps-with-the-openai-api), and [Codex's locked-down computer use on macOS](codex-locked-computer-use-mac).*

## Related reads

- [LifeSciBench: AI Fails 64% of Life-Science Tasks 2026](https://nerdleveltech.com/openai-lifescibench-ai-life-science-benchmark)

[^1]: OpenAI, "Introducing new capabilities to GPT-Rosalind," June 3, 2026. https://openai.com/index/introducing-new-capabilities-to-gpt-rosalind/
[^2]: OpenAI, "Introducing GPT-Rosalind for life sciences research," April 16, 2026 (launch date, customer roster, and model naming). https://openai.com/index/introducing-gpt-rosalind/
[^3]: OpenAI, "Introducing GPT-5.5," April 23, 2026. https://openai.com/index/introducing-gpt-5-5/
[^4]: "DeepMind Spinout Isomorphic Labs Raises $2.1 Billion to Design Drugs With AI," Bloomberg, May 12, 2026. https://www.bloomberg.com/news/articles/2026-05-12/deepmind-spinout-isomorphic-labs-raises-2-1-billion-to-design-drugs-with-ai