OpenAI's GPT-Rosalind: How the Life Sciences Model Is Transforming Research

What is GPT-Rosalind?

GPT-Rosalind is a reasoning model from the GPT family, optimized for scientific workflows. It combines a deep understanding of chemistry, protein engineering, and genomics with an enhanced ability to utilize scientific tools and databases within multi-step processes.

Tasks for which OpenAI positions it:

• Literature Review and Summary of the Evidence
• Interpretation of sequence-function relationships (DNA, RNA, proteins)
• Experimental design and data analysis
• Research on molecules, biological pathways, and disease-related biology
• Multi-step workflow orchestration via the Codex Life Sciences plugin (50+ tools and public databases)

In addition to the model, OpenAI is releasing a Codex Life Sciences plugin on GitHub (free of charge). It provides an orchestration layer for over 50 public resources (human genetics, functional genomics, protein structure, biochemistry, clinical data, and public datasets). This plugin also works with the main GPT-5+ models for users without access to GPT-Rosalind.

Reported performance: results to take seriously

OpenAI has published several benchmarks in which GPT-Rosalind ranks first among the models tested on scientific tasks:

• BixBench (real-world bioinformatics tasks): a score of 0.751, compared to 0.732 for GPT-5.4 and 0.728 for GPT-5. The best result published to date.
• LABBench2 (literature search, database access, sequence manipulation, protocol design): outperformed GPT-5.4 on 6 out of 11 tasks, with the largest improvement observed on CloningQA (design of DNA and enzymatic reagents for molecular cloning protocols).
• Dyno Therapeutics evaluation (RNA sequence-function prediction and generation): top submissions outperformed human experts in prediction by more than 95 percentiles, and ranked around the 84^th percentile in generation.

These figures are compelling. They show that a model specialized in biology outperforms a general-purpose model of the same generation on specific scientific tasks. For teams engaged in large-scale drug discovery, the potential productivity gains are significant.

Access requirements: a model reserved for a select group

This is where the announcement requires careful reading. GPT-Rosalind is being launched through a “secure access program,” with several cumulative conditions:

• Initially available to eligible enterprise customers in the United States — no European availability has been announced as of the date of this announcement.
• OpenAI's safety qualification and assessment: legitimate scientific use, governance, compliance, and abuse prevention.
• Access via ChatGPT, Codex, and the API, subject to certain conditions; the model is not available on a self-service basis like GPT-5.
• Key partners: Amgen, Moderna, Novo Nordisk, Thermo Fisher Scientific, the Allen Institute, NVIDIA, Benchling, Oracle Health, UCSF, and Los Alamos National Laboratory. And consulting partners for integration: McKinsey, BCG, and Bain.

This access strategy—which is justifiable from a biosafety perspective—also shapes the product’s positioning: GPT-Rosalind is aimed at large U.S. pharmaceutical and biotech companies capable of engaging a consulting firm to implement the model. It excludes the majority of European academic laboratories, biotech SMEs, and public research teams.

What this means for European laboratories

For a director of a joint research unit (UMR), an R&D manager in biotechnology, or a team leader in translational oncology in France, several questions arise:

• When will it be available in Europe? No timeline has been announced. Based on OpenAI’s experience with its Enterprise products, it could take anywhere from several months to several quarters before a controlled rollout outside the U.S.
• What does data sovereignty entail? The platform is hosted on OpenAI’s infrastructure (Microsoft Azure in the United States). The U.S. Cloud Act applies. For sensitive research data—unpublished results, patentable hypotheses, patient data—this is a major sticking point for many European organizations.
• What are the operational costs? It’s free during the preview period, but will eventually be subject to a fee. The need to hire a consulting firm for integration increases the total cost.
• What kind of strategic dependency is this? Adopting GPT-Rosalind means basing part of one’s R&D pipeline on a single non-European supplier, in a sector where technological sovereignty is becoming a key industrial issue.

None of these questions is a deal-breaker—but they require a clear answer before committing to critical data and workflows.

Charlie : a sovereign alternative, available today

Charlie, Emerit Science's scientific AI agent, targets the same workflows as GPT-Rosalind—literature search, data analysis, experimental design, and biological reasoning—but with a different approach in three key areas:

• French sovereignty: Scaleway hosting in France, compliant with the GDPR and the AI Act; the Cloud Act does not apply. ProConnect authentication for government employees.
• Included European sources: Charlie natively indexes HAL (French-language scientific output), bioRxiv, PubMed, PMC, and the web. Clickable citations with unified identifiers es_id and cross-index deduplication.
• Available to labs of all sizes: free account, Pro and Team plans, Enterprise quotes. No eligibility requirements, no consulting firm required to get started. A graduate student can try out Charlie in two minutes; a 30-person research unit can set up a shared project in just a few days.

Charlie does not claim to outperform GPT-Rosalind on pure bioinformatics benchmarks. Our strength lies elsewhere: a unified workflow—technological and economic monitoring, data analysis, experimental design, and integrations (eLabFTW, ProConnect, IndexAPI)—available to teams that need to get to work today, within a framework that complies with European regulations.

Comparison Chart

“GPT-Rosalind proves one thing: scientific research deserves dedicated models, not repurposed general-purpose assistants. That’s exactly the conviction that led us to build Charlie. However, a model reserved for major U.S. players doesn’t solve the problem for the majority of European laboratories—and it is this majority that Charlie serves on a daily basis.” — The Emerit Science Team

Who should choose what?

The two tools are not in direct competition: GPT-Rosalind targets a specific segment (U.S. Big Pharma), while Charlie serves the European research ecosystem in all its diversity—from PhD students to R&D executives.

Conclusion: Confirmation that a sovereign scientific AI was necessary

GPT-Rosalind marks a significant milestone: OpenAI has officially validated the idea that an AI model tailored to the life sciences offers a measurable advantage over general-purpose models. This is excellent news for the broader scientific community.

But the announcement also highlights a structural risk for Europe: allowing the next generation of scientific tools to be built exclusively on the infrastructure of a few American companies, with access conditions that exclude mid-sized laboratories and public entities. This is precisely the raison d'être of Charlie: to enable European teams to benefit from advances in scientific AI without sacrificing their data sovereignty or strategic independence.

The question isn’t “GPT-Rosalind or Charlie?” but “What does my lab need right now to move forward, within a framework I can control?” For the vast majority of the teams we support, the answer is clear.