Citation and evidence

GPT-Rosalind

12 min full readUpdated 19 references

This article's verification

Report a problem with this article

Edit article
More

Use this article

Raw MarkdownExplore connections

Improve this page

Revision historyDiscuss this article

Topics

AI ModelsHealthcare AI

Cite this article

GPT-Rosalind is a family of frontier reasoning models for the life sciences, developed by OpenAI and first released on April 16, 2026.[1][6] Named after the British chemist and X-ray crystallographer Rosalind Franklin, whose diffraction work helped reveal the structure of DNA, the models are designed to reason across biology, scientific evidence, data, and tools to accelerate early-stage research in drug discovery, genomics, and translational medicine.[1] OpenAI calls GPT-Rosalind the first release in its life sciences model series, and distributes it not as a general consumer product but through a gated "trusted access" program for organizations with legitimate biology research use cases.[1][3] The series runs in ChatGPT, the OpenAI API, and the company's Codex agentic coding environment, and is paired with a freely available Life Sciences research plugin that connects models to more than 50 scientific tools and databases.[1][7] The April model was followed by GPT-Rosalind-5.5 on June 3, 2026, by its exit from research preview, announced on September 11, 2026, and by GPT-Rosalind-6, announced on October 30, 2026 and sold as "GPT-6 Rosalind Research."[2][17][18] Silicon Republic framed the April launch as OpenAI's answer to Google DeepMind's AlphaFold; it also fits a broader 2026 push by the company into AI for science and AI in healthcare.[15]

Related: GPT-5.5GPT-6 AstraGPT-5.4Grok 4.3

OpenAI logo
OpenAI logo. Source: Wikimedia Commons, Public domain.

Overview

GPT-Rosalind is a specialized large language model optimized for scientific workflows rather than open-ended chat. OpenAI describes it as built "for modern scientific work across published evidence, data, tools, and experiments," with strength in reasoning over molecules, proteins, genes, pathways, and disease-relevant biology.[1] The model targets multi-step research tasks such as target discovery and validation, genomics interpretation, pathway analysis, literature synthesis, sequence-to-function interpretation, experimental planning, and data analysis. OpenAI frames the system as a tool to help scientists "explore more possibilities, surface connections that might otherwise be missed, and arrive at better hypotheses sooner."[1]

The model name honors Rosalind Franklin (1920-1958), whose rigorous X-ray crystallography helped establish the molecular structures of DNA, RNA, and viruses and laid foundations for modern molecular biology. The choice reflects OpenAI's positioning of the model as a tool for rigorous, evidence-driven discovery. GPT-Rosalind followed earlier OpenAI scientific efforts, including the Prism scientific writing platform launched in January 2026, and arrived amid a wave of generative AI products aimed at the research enterprise.

AttributeDetail
DeveloperOpenAI
Initial releaseApril 16, 2026 (research preview)[1][6]
GPT-Rosalind-5.5June 3, 2026; incrementally trained from GPT-5.5[2][19]
GPT-Rosalind-6October 30, 2026; same weights as GPT-6 Astra, sold as "GPT-6 Rosalind Research"[17][18]
Model typeFrontier reasoning models for life sciences
NamesakeRosalind Franklin, DNA-structure crystallographer
DomainsDrug discovery, genomics, protein engineering, translational medicine
AvailabilityChatGPT, API, and Codex through the trusted-access program; out of research preview since September 11, 2026[2][3]
Access tiersRosalind Research and Rosalind Discovery, plus the Rosalind Academic Plan for U.S. university labs[3]
Preparedness classificationHigh capability in the Biological and Chemical domain, below Critical (both GPT-Rosalind-5.5 and GPT-Rosalind-6)[17][19]
Companion toolFree Life Sciences research plugin for Codex (50+ tools and databases)[7]
Launch partnersAmgen, Moderna, the Allen Institute, Thermo Fisher Scientific[1]
Additional collaboratorsDyno Therapeutics, Los Alamos National Laboratory, Novo Nordisk[1][2]
Research leadJoy Jiao (Life Sciences Research Lead)[14]
Product leadYunyun Wang (Life Sciences Product Lead)[14]
Biodefense programRosalind Biodefense (launched May 29, 2026)[13]

Expanded article table

Announcement and partners

OpenAI introduced GPT-Rosalind on April 16, 2026, publishing a dedicated announcement and an OpenAI Help Center article describing the model and its access program; Axios reported the launch the same day and other trade outlets followed over the next week.[1][3][6][8][10][11][12][16] The company named launch customers Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific, and quoted executives from Oracle Health and Life Sciences, NVIDIA, and Benchling, plus a University of California, San Francisco pharmacy professor, alongside them.[1][5] OpenAI also cited a collaboration with Dyno Therapeutics and ongoing work at Los Alamos National Laboratory on AI-guided protein and catalyst design.[1] The launch came the same week OpenAI announced a separate strategic alliance with Novo Nordisk, whose chief executive Mike Doustdar was quoted in the announcement, underscoring a concerted move into pharmaceutical and biotech partnerships.[1]

Industry partners emphasized the model's fit for high-stakes, data-intensive research. Sean Bruich, Amgen's senior vice president of artificial intelligence and data, said the life sciences field "demands precision at every step," noting that "the questions are highly complex, the data are highly unique, and the stakes are incredibly high."[1] After the June update, product lead Yunyun Wang told R&D World that users could expect higher performance on life sciences research tasks with GPT-Rosalind and more consistent results when it is combined with the Codex plugins as an execution and orchestration layer, while research lead Joy Jiao said the plugin workflows also run well with GPT-5.5.[14] OpenAI argued that gains made at the earliest stages of discovery compound downstream in better target selection, stronger biological hypotheses, and higher-quality experiments, a core part of the value proposition for AI drug discovery.[1]

Access is deliberately restricted. At launch GPT-Rosalind went to qualified Enterprise customers in the United States first, behind a qualification and safety review.[1][9] With the June 3, 2026 update OpenAI expanded access to eligible organizations globally and added an OpenAI-managed workspace for qualified organizations without an Enterprise account.[2] On September 11, 2026 it said the model was coming out of research preview and was available globally to eligible organizations through the trusted-access program, with published pricing taking effect on October 5, 2026.[1][2] As of 31 October 2026 the Help Center splits access into two reviewed tiers: Rosalind Research for frontier biological work, which covers GPT-5.5 Rosalind Research and GPT-6 Rosalind Research, and Rosalind Discovery for everyday scientific work on the Rosalind 5.5 model under stricter monitoring; approval for Discovery does not grant Research access.[3] Academic labs at U.S. universities can apply separately through the Rosalind Academic Plan, which does not include API access or API credits, and the model is not offered through Amazon Bedrock.[3] OpenAI also presents Rosalind as a brand rather than a single model: Rosalind Workbench is a scientific workspace in Codex and ChatGPT with next-generation sequencing workflows for FASTQ quality control and bulk and single-cell RNA sequencing, and it does not itself require GPT-Rosalind model access.[3][4] Alongside the model, OpenAI released a free Life Sciences research plugin for Codex that connects to more than 50 scientific databases and tools, available more broadly than the gated model itself.[1][7]

Capabilities

In its own evaluations, OpenAI said at launch that GPT-Rosalind delivered the best performance on tasks requiring reasoning over molecules, proteins, genes, pathways, and disease-relevant biology; its core-benchmark chart compared it only with earlier OpenAI models.[1] On BixBench, a benchmark built around real-world bioinformatics and data analysis, the model posted 0.751, which OpenAI described as leading performance among models with published scores; its chart put GPT-5.4 at 0.732 and also showed Grok 4.2 and Gemini 3.1 Pro.[1] On LABBench2, a benchmark covering literature retrieval, database access, sequence manipulation, and protocol design, OpenAI reported that GPT-Rosalind outperformed GPT-5.4 on 6 of 11 tasks, with the largest gain on CloningQA, which requires end-to-end design of DNA and enzyme reagents for molecular cloning protocols.[1] OpenAI's safety documentation separately cites the 2024 LAB-Bench paper for the ProtocolQA protocol-troubleshooting dataset used in its Preparedness evaluations.[17] With Dyno Therapeutics, OpenAI also tested the model on an RNA sequence-to-function prediction and generation task built from unpublished, uncontaminated sequences and compared it against 57 historical scores from human experts in the AI-bio field: evaluated in the Codex app, best-of-ten submissions ranked above the 95th percentile of those experts on prediction and around the 84th percentile on sequence generation.[1]

On June 3, 2026, OpenAI announced a model update that was incrementally trained from GPT-5.5, combining that model's agentic coding and tool use with stronger domain intelligence in core drug-discovery areas such as medicinal chemistry and genomics.[2][19] The June announcement's headline and body call the update simply GPT-Rosalind (its page title now reads "Introducing GPT-Rosalind-5.5"); the system card and the later addendum name it GPT-Rosalind-5.5, and customers select it as "GPT-5.5 Rosalind Research."[2][3][17][19] OpenAI introduced new evaluations alongside the update: GeneBench for long-horizon, end-to-end genomics and quantitative-biology analysis, MedChemBench for medicinal chemistry, LifeSciBench for applied research tasks judged by outside experts, and LabWorkBench for wet-lab troubleshooting built on proprietary protocols.[2] All of the figures below are self-reported.

Evaluation (June 2026 version)GPT-Rosalind-5.5GPT-5.5Token use
MedChemBench27.5%25.1%7.2% fewer tokens[2]
GeneBench21.6%20.4%31% fewer tokens[2]
LabWorkBench63.2%55.8%5.3% fewer tokens[2][19]
LifeSciBench (overall)63.4%58.8%34,439 vs 33,351 average output tokens (more, not fewer)[2][19]

Expanded article table

On the overall LifeSciBench score OpenAI put GPT-Rosalind-5.5 ahead of GPT-5.5, Gemini 3.1 Pro at 51.7 percent and Grok 4.3 at 33.7 percent.[2] Its system card for the update reported High capability in the Biological and Chemical domain under the Preparedness Framework while falling below the Critical threshold, with cybersecurity capability at or below that of GPT-5.5.[19]

The model's intended workflow is agentic and tool-augmented: rather than answering single questions, it is designed to plan and execute multi-step research, calling external scientific tools and databases through the Codex plugins to interpret sequences, analyze pathways, review literature, and propose experiments. The June update added a second plugin, Life Sciences NGS Analysis, for repeatable bioinformatics execution, plus interactive sequence, alignment, and structure viewers in Codex.[2] This positions GPT-Rosalind as a reasoning layer over the scientific software stack rather than a replacement for specialized structure-prediction or simulation systems.

GPT-Rosalind-6

OpenAI announced GPT-Rosalind-6 on October 30, 2026 and published a system card addendum for it the same day.[17][18] The model uses the same weights as GPT-6 Astra, OpenAI's flagship reasoning model, with no added training data and no further post-training for the release; what differs is the deployment, which allows fewer refusals on dual-use biological queries and is available only through trusted access.[17] Customers see it as "GPT-6 Rosalind Research" in ChatGPT Work and Codex, not in the Chat tab, for a limited set of approved Enterprise organizations; in Enterprise workspaces it begins with a two-week Admin Preview that is off by default, the API model identifier is gpt-6-rosalind where an organization is separately approved, and GPT-5.5 Rosalind Research stays selectable for existing workflows.[3][18] Under the Preparedness Framework OpenAI again treats the launch as High capability in the Biological and Chemical domain and below the Critical threshold, relies on GPT-6 Astra's assessments for the other tracked risks, and treats the model as Critical in Cybersecurity and below High in AI self-improvement.[17] At the highest reasoning budget tested, the addendum reported 49.6 percent mean rubric score on MedChemBench v7 against 28.9 percent for GPT-Rosalind-5.5, 37.7 percent single-attempt success on GeneBench Pro v13 against 16.7 percent, and 61.3 percent on LifeSciBench Gold against 44.0 percent; OpenAI notes that these GPT-Rosalind-5.5 numbers use the same benchmark versions and problem sets as GPT-Rosalind-6 and therefore differ from the ones reported at its own launch, so the two sets should not be mixed.[17] A fuller account is in GPT-Rosalind-6.

Significance

GPT-Rosalind was a step by a leading foundation-model developer into specialized scientific AI, extending the reach of general-purpose generative AI into the regulated, high-stakes world of biomedical research. By gating the model behind a trusted-access program and pairing it with vetted enterprise partners, OpenAI signaled an approach that treats frontier biological capability as dual-use, requiring qualification and safety review rather than open release. Commentators noted that the model could compress timelines in early drug discovery, though access restrictions meant most researchers could not use it directly at launch.[9]

The biosecurity dimension became explicit on May 29, 2026, when OpenAI launched Rosalind Biodefense, an initiative for defensive applications of AI in the life sciences.[13] The program operates two tracks: one sponsoring vetted outside developers building pandemic-preparedness and biosecurity tools (in areas such as epidemiological modeling, early detection, screening, and non-pharmaceutical interventions), and another providing access to select U.S. government and allied partners.[13] R&D World reported OpenAI's on-background statement that it had briefed the White House and several federal agencies on the approach, and OpenAI named collaborators including Lawrence Livermore National Laboratory (pairing the model with supercomputing to design and evaluate medical countermeasures), the Johns Hopkins Applied Physics Laboratory (integrating GPT-Rosalind into protein-engineering platforms to screen mutant enzymes), and the Coalition for Epidemic Preparedness Innovations (applying it to vaccine development, including work related to a Bundibugyo Ebola outbreak in the Democratic Republic of the Congo and Uganda).[13] OpenAI framed the effort as "defensive acceleration," arguing that frontier AI "should meaningfully advantage those defenders."[13] The announcement drew attention in part because it came shortly after the U.S. administration postponed an executive order that would have created government review processes for powerful AI models before release.[13]

Taken together, the launch and its follow-ups positioned GPT-Rosalind as a flagship of OpenAI's 2026 AI for science strategy, a competitor in scientific AI alongside structure-prediction systems, and a test case for how frontier AI in healthcare and biology can be deployed under access controls.

Continue reading

More to read

Topics:AI ModelsHealthcare AI

References

  1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21 ^22OpenAI, "Introducing GPT-Rosalind for life sciences research," openai.com/...introducing-gpt-rosalind
  2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15OpenAI, "Introducing new capabilities to GPT-Rosalind," openai.com/...ing-new-capabilities-to-gpt-rosalind
  3. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9OpenAI Help Center, "GPT-Rosalind for life sciences research," help.openai.com/...lind-for-life-sciences-research
  4. ^OpenAI, "Rosalind," openai.com/rosalind
  5. ^OpenAI on X, partner announcement (Amgen, Moderna, Allen Institute, Thermo Fisher), x.com/...2044861695911477643
  6. ^1 ^2 ^3Axios, "OpenAI launches new AI model for life sciences research," axios.com/...openai-models-life-sciences-drugs
  7. ^1 ^2 ^3VentureBeat, "OpenAI debuts GPT-Rosalind, a new limited access model for life sciences, and broader Codex plugin on GitHub," venturebeat.com/...-broader-codex-plugin-on-github
  8. ^FierceBiotech, "OpenAI launches biotech-specific AI model dubbed GPT-Rosalind," fiercebiotech.com/...pecific-ai-model-gpt-rosalind
  9. ^1 ^2Decrypt, "OpenAI's New AI Model Rosalind Could Shave Years Off Drug Discovery. You Probably Can't Use It," decrypt.co/...d-drug-discovery-life-sciences-model
  10. ^Euronews, "What to know about OpenAI's new model for life sciences research GPT-Rosalind," euronews.com/...ife-sciences-research-gpt-rosalind
  11. ^pharmaphorum, "OpenAI introduces GPT-Rosalind, its drug discovery AI," pharmaphorum.com/...rosalind-its-drug-discovery-ai
  12. ^TechTarget, "OpenAI debuts AI model GPT-Rosalind to speed up drug discovery," techtarget.com/...alind-to-speed-up-drug-discovery
  13. ^1 ^2 ^3 ^4 ^5 ^6R&D World, "OpenAI launches Rosalind Biodefense, offers federal agencies early access to its life-sciences model," rdworldonline.com/...ss-to-its-life-sciences-model
  14. ^1 ^2 ^3R&D World, "OpenAI research and product leads detail GPT-Rosalind capabilities and benchmarks," rdworldonline.com/...d-capabilities-and-benchmarks
  15. ^Silicon Republic, "OpenAI to rival Google's AlphaFold with new AI model for life sciences research," siliconrepublic.com/...odel-life-sciences-research
  16. ^Lab Manager, "OpenAI Launches GPT-Rosalind to Accelerate Research and Drug Discovery," labmanager.com/...esearch-and-drug-discovery-35273
  17. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9OpenAI, "GPT-Rosalind-6 System Card Addendum," Deployment Safety Hub, published October 30, 2026, deploymentsafety.openai.com/gpt-rosalind-6
  18. ^1 ^2 ^3 ^4OpenAI Help Center, "ChatGPT Enterprise and Edu release notes," entry for October 30, 2026, help.openai.com/...nterprise-and-edu-release-notes
  19. ^1 ^2 ^3 ^4 ^5 ^6 ^7OpenAI, "GPT-Rosalind-5.5 System Card," Deployment Safety Hub, published June 3, 2026, deploymentsafety.openai.com/gpt-rosalind-5-5

Improve this article

Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.

7 revisions · v7 · 2,485 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here

Reviewer note: Independently verified 2026-10-31 (see v6 stamp); v7 only attaches existing launch-coverage references, whose April 17-22, 2026 publication dates were checked.

Cite this page: AI Wiki. "GPT-Rosalind." aiwiki.ai, updated 31 Oct 2026, fact-checked 31 Oct 2026. CC BY 4.0. https://aiwiki.ai/wiki/gpt_rosalind

Edit article