OpenAI Launches Rosalind Workbench: AI-Powered Research Platform Aims to Put an Entire Life-Sciences Research Team in One Workspace

OpenAI has introduced Rosalind Workbench, a new research-preview platform designed to bring artificial intelligence, scientific databases, biological data, specialised models and research tools together in a single workspace. The company says the platform is aimed at helping scientists move more efficiently from a research question and raw data to evidence, analysis and the planning of the next experiment.

The new platform is built around GPT-Rosalind, OpenAI’s specialised AI model for life-sciences research. Unlike a general-purpose chatbot, GPT-Rosalind has been developed specifically for workflows involving biology, genomics, drug discovery, proteins, medicinal chemistry and experimental research.

What is Rosalind Workbench?

Rosalind Workbench is essentially a digital scientific research environment.

In traditional research, a scientist may need to move between multiple systems—one for sequencing data, another for analysing proteins, another for scientific literature, another for molecular structures and yet another for documenting experiments.

OpenAI’s new approach is to connect many of these activities inside one workflow.

The company describes Rosalind Workbench as an “orchestrated workspace” where researchers can combine AI reasoning, scientific tools, data and evidence while keeping the research process reviewable.

The platform is currently available as a research preview, meaning OpenAI is continuing to develop and evaluate it with researchers and organisations.

GPT-Rosalind: The AI engine behind the platform

At the centre of the ecosystem is GPT-Rosalind, a purpose-built model for life-sciences research.

OpenAI first introduced GPT-Rosalind in April 2026 and has since expanded its capabilities. The model is designed to help with tasks including:

• Biological evidence synthesis
• Genomics and sequencing analysis
• Protein and sequence analysis
• Medicinal chemistry
• Target discovery and validation
• Literature research
• Hypothesis generation
• Experimental planning
• Wet-lab troubleshooting
• Analysis of complex biological data

The important difference is that GPT-Rosalind is not simply being positioned as an AI that answers scientific questions. It is designed to work with scientific tools and data as part of a multi-step research workflow.

From raw biological data to the next experiment

One of the biggest changes introduced through Rosalind Workbench is the attempt to connect different stages of scientific research.

For example, a research project could begin with a sequencing dataset.

The workflow could look something like:

Sequencing data → Quality control → Analysis → Biological interpretation → Evidence review → Hypothesis → Experiment planning

Instead of treating each stage as an isolated task, Rosalind Workbench is designed to keep the steps connected.

Researchers can bring sequencing files and sample information into a workflow, review the proposed analysis plan, follow the analysis, inspect quality-control metrics and examine saved outputs before deciding what biological question to investigate next.

Genomics and NGS analysis

One of the practical capabilities highlighted by OpenAI is next-generation sequencing (NGS) analysis.

The Workbench supports guided workflows involving:

• FASTQ quality control
• Bulk RNA sequencing
• Single-cell RNA sequencing
• Genomic data analysis
• Biological interpretation

This means researchers can begin with actual sequencing data rather than simply asking a general AI model a question about genomics.

The system is designed to allow researchers to review inputs and analysis plans before proceeding, while preserving outputs that can be examined later.

Understanding proteins and biological structures

Another major area is protein research.

Rosalind Workbench can bring together protein sequences, sequence alignments and molecular structures, allowing researchers to examine them alongside AI reasoning.

This could be useful when investigating questions such as:

• How might a genetic variant affect a protein?
• Which part of a protein could be involved in an interaction?
• How might changes in a sequence influence protein function?
• Which experiment could help validate the hypothesis?

OpenAI says the platform includes interactive scientific viewers so that researchers can remain close to the underlying biological evidence while working with the AI.

Drug discovery and medicinal chemistry

Rosalind also targets one of the most commercially important areas of life-sciences research: drug discovery.

The system can help researchers compare molecular candidates, examine molecular properties, consider structure–activity relationships and combine those findings with published evidence.

A simplified workflow could be:

Disease → Biological target → Molecular candidates → Evidence → Candidate comparison → Experimental validation

The goal is not to replace laboratory testing. Instead, AI is intended to help researchers analyse more information and identify promising questions or candidates before experiments are conducted.

Scientific literature and evidence synthesis

Another important capability is connecting scientific literature with experimental data.

Researchers often need to compare information from:

• Scientific papers
• Biological databases
• Existing experimental results
• Genomic data
• Protein information
• Internal research records

Rosalind Workbench is designed to help researchers bring these sources together, identify where evidence agrees or conflicts and determine what question should be investigated next.

This could potentially reduce one of the biggest problems in modern research: important information being scattered across different tools and sources.

New scientific viewers

OpenAI has also introduced specialised viewers that allow scientists to interact with biological information rather than viewing everything as plain text.

The platform includes tools for areas such as:

Molecular Structure Viewer

For examining three-dimensional molecular and protein structures.

Biological Sequence and Alignment Viewer

For comparing biological sequences and alignments.

Slide Viewer

For working with biological slide data.

These viewers are intended to work alongside the AI so researchers can ask questions about the information they are currently examining.

Two ways to use Rosalind Workbench

The platform provides two broad modes.

Explore Mode

This is intended for general scientific questions and exploration using the latest ChatGPT models available to the user.

Research Mode

This is designed for more complex biological questions and advanced research workflows using GPT-Rosalind.

Research-mode access is currently subject to eligibility and organisational approval.

What makes this different from normal ChatGPT?

This is perhaps the most important question.

A normal AI chatbot might answer:

“What genes are associated with this disease?”

Rosalind is designed to go further.

A researcher could potentially ask a broader research question and then move through:

Evidence → Biological data → Analysis → Interpretation → Hypothesis → Experiment

The distinction is therefore not simply a smarter chatbot.

It is an attempt to create an AI-assisted research environment where models, databases, scientific software, visualisation tools and experimental workflows can work together.

OpenAI’s benchmark results

OpenAI says GPT-Rosalind has been evaluated on several life-sciences research benchmarks.

The company reports improvements in performance per token of:

• 53.7% on Genebench
• 18.0% on MedChem Bench
• 19.6% on Labworkbench
• 4.42% on LifeSci Bench

These are OpenAI-reported results, so they should be interpreted as company evaluation results rather than independent confirmation of real-world scientific impact.

Who is working with OpenAI?

OpenAI has been working with several organisations in the life-sciences ecosystem.

Its announced ecosystem and partner references include companies and institutions such as:

• Novo Nordisk
• Thermo Fisher Scientific
• Moderna
• Oracle Health and Life Sciences
• Amgen
• NVIDIA
• Allen Institute
• Benchling
• UCSF School of Pharmacy

These collaborations are important because the ultimate test of an AI research system is not simply benchmark performance—it is whether researchers can actually use it in real scientific workflows.

Why this could matter for drug discovery

Drug discovery is traditionally a long and expensive process.

Researchers must identify potential biological targets, understand disease mechanisms, find promising molecules, test them and repeatedly refine their hypotheses.

AI could potentially shorten some of the early stages by helping researchers:

Search → analyse → compare → hypothesise → design experiments → evaluate results

The laboratory still remains essential, but AI could help researchers reach better-informed experimental decisions faster.

OpenAI’s bigger vision

OpenAI says Rosalind is designed around the idea that an individual scientist could effectively have access to a much broader set of research capabilities.

The company’s vision is to connect:

AI reasoning + scientific tools + databases + experimental data + evidence

into reusable research workflows.

Over time, OpenAI says it envisions teams of AI agents working across different scientific domains, potentially giving researchers access to broader expertise and enabling more ambitious research questions.

What about safety?

Because Rosalind deals with advanced biological capabilities, OpenAI is also positioning the platform around controlled access.

GPT-Rosalind is currently available through a trusted-access model for eligible organisations, with governance and security requirements. OpenAI also describes a separate Rosalind Biodefense initiative focused on defensive applications such as early detection, preparedness, diagnostics, response and medical countermeasures.

What comes next?

Rosalind Workbench is still in research preview, so the platform should not be considered a finished replacement for scientific laboratories or expert researchers.

Its significance is instead in the direction it represents.

OpenAI is moving from:

AI that answers questions

towards:

AI that participates in structured scientific workflows.

If the technology develops as intended, future versions could help scientists move from a biological question to data analysis, evidence synthesis, hypothesis generation and experiment planning within a much more connected environment.

The bigger picture

The launch of Rosalind Workbench represents another step in the rapidly developing relationship between AI and scientific research.

For scientists, the potential benefit is not simply getting faster answers. It is having a system that can help connect data, evidence, scientific tools and reasoning throughout the research process.

And that could ultimately change how some early-stage biological research and drug-discovery projects are conducted.