ModelsArchitectures & capability
OpenAI offers free GPT-Rosalind seats to US university labs through new Rosalind Academic Plan
OpenAI's Rosalind Academic Plan gives eligible US university labs up to 30 free ChatGPT Business Standard seats for up to a year. Using its life-sciences model GPT-Rosalind still requires a separate trusted-access review that names a university biosafety officer.
From the OpenAI News article of October 15, 2026 and cited sources: Eligible labs get up to 30 free ChatGPT Business Standard seats for up to 12 months, with ChatGPT and Codex included under existing usage limits. [1] [2] [9]
SOURCES[1] Bringing GPT-Rosalind to academic research labs (OpenAI) openai.com[2] OpenAI Launches Rosalind Academic Plan for U.S. University… unite.ai[9] Rosalind Academic Plan (OpenAI Help) help.openai.com
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- Eligible labs get up to 30 free ChatGPT Business Standard seats for up to 12 months, with ChatGPT and Codex included under existing usage limits. [1] [2] [9]
- Original source
- OpenAI News article, company source
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- October 15, 2026 article published
- Covered
- Oct 18 edition
- Evidence
- 13 sources across 11 sites
- 1Bringing GPT-Rosalind to academic research labs (OpenAI)openai.com
- 2OpenAI Launches Rosalind Academic Plan for U.S. University…unite.ai
- 3OpenAI's biology model is an access strategy (Implicator)implicator.ai
- 4OpenAI launches GPT-Rosalind (The Decoder)the-decoder.com
- 5OpenAI built a specialty model for drug discovery…letsdatascience.com
- 6OpenAI launches Rosalind Biodefense (R&D World)rdworldonline.com
QuietHow it’s measured
Hacker News · 3 likes
Reach blends public engagement (views, likes, comments, shares, stars) from the platforms where the story or its original source lives: YouTube, Hacker News, GitHub, Substack and similar. A video and the article written from its transcript share one score, taken from the video. Platforms with no data are left out rather than counted as zero, and the age of a story never lowers its score.
Eligible labs get up to 30 free ChatGPT Business Standard seats for up to 12 months, with ChatGPT and Codex included under existing usage limits. The principal investigator applies. Members need .edu addresses on the PI's domain, so outside collaborators can't join, and Zero Data Retention is not offered. [1] [2] [9]
The academic plan continues a staged rollout. GPT-Rosalind opened in April 2026 as a research preview for vetted enterprise partners such as Amgen and Moderna. A biodefense program followed in May with groups including the WHO and Lawrence Livermore. A June update built on GPT-5.5 then opened trusted access to eligible organizations worldwide. [4] [6] [7]
On BixBench, GPT-Rosalind scored 0.751 pass@1, compared with 0.732 for GPT-5.4 and 0.550 for Gemini 3.1 Pro. Because OpenAI reported these figures itself, the roughly two-point lead over GPT-5.4 is small enough that it could change if someone else reran the test. [4] [5]
Critics make two arguments. Vetting who gets access protects little once many staff at an approved institution can query the model, and screening DNA synthesis orders matters more. Limiting access to established institutions may also leave smaller labs behind. [5]
What is documented is a lower cost: free seats remove a price barrier for academic labs. What is not yet shown is whether academic use of a gated biology model will grow, since safety review, not budget, decides which labs get in.
Executive brief
Starting October 15, 2026, OpenAI is giving US university labs its restricted life-sciences model for free. Until then, GPT-Rosalind had gone mostly to large drug companies and selected government partners. Under the new Rosalind Academic Plan, an eligible lab gets up to 30 ChatGPT Business Standard seats at no cost for up to 12 months (Unite.AI). GPT-Rosalind itself is only switched on after a separate "trusted-access" review, and each application must name a biosafety officer from the university. The free seats lower the cost barrier, but the safety review still decides who gets in. One commentator's view: OpenAI's real product here is controlled access, not a scientific breakthrough (Implicator).
What changed and event timeline
GPT-Rosalind launches as a research preview
Access was limited to vetted US enterprise customers. Named partners included Amgen, Moderna, Novo Nordisk, Thermo Fisher, NVIDIA, the Allen Institute and UCSF School of Pharmacy (;).
Rosalind Biodefense launches
Vetted developers and selected government and public-health bodies get early access for defensive biosecurity work. Named participants include Lawrence Livermore National Laboratory, Johns Hopkins APL, CEPI and the WHO ().
Model update built on GPT-5.5
OpenAI reports gains in medicinal chemistry, genomics and wet-lab troubleshooting, plus new research plugins. Availability expands to eligible organizations worldwide through trusted access (;).
Rosalind Academic Plan opens
US university labs can apply for free seats, with GPT-Rosalind access depending on trusted-access approval (;).
Capabilities and access
- Model: GPT-Rosalind, a reasoning model for biology, drug discovery and translational medicine. The June version adds GPT-5.5's agentic coding and tool use (OpenAI).
- Academic terms: Only US university labs qualify. ChatGPT and Codex are included under existing usage limits. Zero Data Retention is not available (Unite.AI).
- Universities with an OpenAI Enterprise agreement must apply through their institution instead (Wes Roth on X).
Read the full section
- Model: GPT-Rosalind, a reasoning model for biology, drug discovery and translational medicine. The June version adds GPT-5.5's agentic coding and tool use (OpenAI). One outlet names an API model "gpt-rosalind-discovery" priced at $5 per million input tokens; no OpenAI source in this review confirms that (Crypto Briefing).
- Academic terms: Only US university labs qualify. The PI applies, and all members need .edu emails on the PI's domain, so visiting researchers and outside collaborators can't join. ChatGPT and Codex are included under existing usage limits. Zero Data Retention is not available (Unite.AI).
- Universities with an OpenAI Enterprise agreement must apply through their institution instead (Wes Roth on X).
Technical analysis for researchers and developers
- Tooling: At launch, a free life-sciences plugin on GitHub connected the model to 50+ public databases and tools, including AlphaFold and BindingDB (The Decoder).
- Evaluation: OpenAI built its own benchmark, LifeSciBench.
- Reproducibility: Because access is gated, outside researchers can't rerun the published scores.
Read the full section
- Tooling: At launch, a free life-sciences plugin on GitHub connected the model to 50+ public databases and tools, including AlphaFold and BindingDB (The Decoder). In June, OpenAI added two plugins, Life Sciences Research and Life Sciences NGS Analysis, for evidence retrieval and bioinformatics execution (OpenAI).
- Evaluation: OpenAI built its own benchmark, LifeSciBench. It is graded by outside experts and covers six workflow areas, from evidence handling to translation and communication (OpenAI).
- Reproducibility: Because access is gated, outside researchers can't rerun the published scores. OpenAI's own vetting also does not check whether proposed methods actually work (Unite.AI).
Claims and evidence
- BixBench: Pass@1 of 0.751, versus 0.732 for GPT-5.4 and 0.550 for Gemini 3.1 Pro (The Decoder).
- LABBench2: Beat GPT-5.4 on 6 of 11 tasks (Implicator).
- Dyno Therapeutics test: The best of ten submissions ranked above the 95th percentile on prediction tasks and around the 84th on sequence generation (Implicator).
Read the full section
- BixBench: Pass@1 of 0.751, versus 0.732 for GPT-5.4 and 0.550 for Gemini 3.1 Pro (The Decoder). These are vendor-reported, and the roughly 2-point lead over GPT-5.4 is within the range that can change when someone else reruns the test (Let's Data Science).
- LABBench2: Beat GPT-5.4 on 6 of 11 tasks (Implicator).
- Dyno Therapeutics test: The best of ten submissions ranked above the 95th percentile on prediction tasks and around the 84th on sequence generation (Implicator).
- Efficiency: Used 31% fewer tokens than GPT-5.5 on genomics analysis while improving accuracy (vendor-reported) (OpenAI).
- Unconfirmed: Gains of 53.7% on GeneBench, 18.0% on MedChem Bench and 19.6% on LabWorkBench appear in only one secondary outlet (Crypto Briefing).
- No independent evaluation of GPT-Rosalind turned up in the reviewed sources.
Context and prior work
- OpenAI rated its July 2025 ChatGPT Agent "high capability" in biology and released it under its Preparedness Framework.
- In February 2026, more than 100 researchers endorsed a framework for restricting access to sensitive biological datasets (Implicator).
- The academic plan follows the same tiered pattern as earlier releases: pharma partners first, then biodefense users, then universities.
Read the full section
- OpenAI rated its July 2025 ChatGPT Agent "high capability" in biology and released it under its Preparedness Framework. It has also reported a 40% cut in cell-free protein-synthesis costs from work with Ginkgo Bioworks (R&D World).
- In February 2026, more than 100 researchers endorsed a framework for restricting access to sensitive biological datasets (Implicator).
- The academic plan follows the same tiered pattern as earlier releases: pharma partners first, then biodefense users, then universities.
Limitations, safety and contested findings
- Safeguards: Access is vetted by institution.
- Critics cited by Let's Data Science argue that controls at the access layer protect little once many employees can query the model, and that screening of DNA synthesis orders matters more (Let's Data Science).
- Benchmarks vs. outcomes: Good benchmark scores don't show that the model picks better drug targets or gets more drugs approved (Implicator).
Read the full section
- Safeguards: Access is vetted by institution. The model is reported to refuse help with bioweapon precursors and gain-of-function work, and applicants must name a biosafety officer (AI Learning Guides).
- Critics cited by Let's Data Science argue that controls at the access layer protect little once many employees can query the model, and that screening of DNA synthesis orders matters more (Let's Data Science).
- Benchmarks vs. outcomes: Good benchmark scores don't show that the model picks better drug targets or gets more drugs approved (Implicator).
- Who is left out: Gating access to established institutions may disadvantage smaller labs (Let's Data Science).
Business and practitioner implications
- For PIs: Free seats for 12 months, but expect paperwork.
- For OpenAI: The plan brings in biology-specific feedback from universities, and graduate students who build workflows on it may carry the habit into pharma and biotech jobs (AI Learning Guides).
- For data owners: Accounts are not used for training by default, but standard data retention applies.
Read the full section
- For PIs: Free seats for 12 months, but expect paperwork. Plan for the biosafety-officer requirement and for the fact that outside collaborators can't be added (Unite.AI).
- For OpenAI: The plan brings in biology-specific feedback from universities, and graduate students who build workflows on it may carry the habit into pharma and biotech jobs (AI Learning Guides).
- For data owners: Accounts are not used for training by default, but standard data retention applies. Labs that need Zero Data Retention or strict privacy for sensitive data should check the terms first.
- For competitors and labs choosing tools: Without independent evaluations, test the model on your own tasks before relying on it.