On October 30, 2026 OpenAI announced GPT-6 Rosalind Research, a variant of its GPT-6 series tuned for multi-step biological and chemical reasoning. The headline detail, buried in the system card addendum, is that the model uses the exact same weights as GPT-6 Astra, OpenAI's general flagship released in early September. No additional training data or post-training was added for this release [1][2][3].
What changes is the refusal boundary and who gets through the door.
What the model is and isn't #
OpenAI describes GPT-6 Rosalind Research as "a frontier reasoning model for biology, drug discovery, and translational medicine" designed for tasks such as genomics, protein and sequence analysis, medicinal chemistry, literature synthesis, wet-lab troubleshooting, and experiment planning [1][2][4]. The announcement does not claim the model can independently make discoveries or replace experimental work [1].
The model is the third generation in the Rosalind series, which launched in April. Its predecessor, GPT-Rosalind-5.5, was built on GPT-5.5. The new version simply swaps in the GPT-6 Astra base [3].
Benchmarks: internal, small, and not about outcomes #
OpenAI publishes three internal benchmarks comparing GPT-Rosalind-6 against GPT-Rosalind-5.5 at the highest reasoning budget:
- MedChemBench v7 (medicinal chemistry, 99 tasks): 49.6% vs 28.9%
- GeneBench Pro v13 (genomics, 129 tasks, single attempt, test configuration differs from production): 37.7% vs 16.7%
- LifeSciBench Gold (applied life sciences research, 94 tasks): 61.3% vs 44.0% [2][3]
These are OpenAI's own tests of roughly a hundred tasks each. They are not independently reproduced. A GeneBench score measures whether an analysis is done correctly, not whether a drug candidate will work. The improvement over Rosalind-5.5 comes from the new Astra base, not from life-sciences-specific training [3].
OpenAI's original GPT-Rosalind announcement in April 2026, updated in September, cited a Dyno Therapeutics evaluation on an RNA sequence-to-function task using unpublished sequences. Best-of-ten model submissions ranked above the 95th percentile of 57 human experts on prediction and around the 84th percentile on sequence generation [4]. That evaluation was for the first-generation Rosalind model (built on GPT-5.5). The provided October 30 sources do not mention a repeat of this evaluation.
Safety ratings: High on bio/chem, Critical on cyber #
The system card rates GPT-6 Rosalind High for biological and chemical capability, below the Critical threshold. On a Biorisk Knowledge test (consensus-at-32) the model scored 78.3% unadjusted and 85.0% refusal-adjusted, above the 80% indicative High threshold, with a 9.4% refusal rate [2].
On the flip side, the model answered more dual-use biology questions. On a set of benign biology questions Rosalind answered fully in 100% of cases against 91.6% for Astra. On sensitive biological material, Rosalind's answers passed OpenAI's safety grader 79.0% of the time against 99.8% for Astra [3]. That is exactly why access is gated.
None of the four biological design evaluations reached the Critical threshold: AAV capsid packaging prediction (developed by Dyno Therapeutics), SHP2 protein function prediction, a coronavirus-ACE2 cell-entry screen, and phage-plasmid co-evolution [2][3].
For cybersecurity the model inherits Astra's Critical rating, with what the card calls a "more conservative cyber refusal boundary" layered on top [2]. Biosecurity red-teaming was conducted on GPT-Rosalind-5.5 before launch and is being repeated for GPT-6 Rosalind [2].
Access is the product #
GPT-6 Rosalind Research is available only to a limited set of approved research customers through Work and Codex, and via API where separately approved. It is not available in the regular ChatGPT chat tab [1][5].
Non-government applicants undergo business verification and compliance screening; government applicants follow a separate trusted-access pathway for biosecurity, biosafety, public-health preparedness, and biodefense. Enterprise customers must apply "qualified human review before outputs inform experiments, publications, or decisions" [2]. The help centre describes broader access requirements for approved organisations [5].
In Enterprise workspaces the model starts in a two-week Admin Preview and is off by default. Existing Research approval does not bypass Admin Preview [5]. The Rosalind Academic Plan, with up to 30 free ChatGPT seats for up to twelve months, is currently open only to labs at US universities [3][5].
OpenAI monitors use after approval and can restrict or revoke access [3]. The company says it evaluates access based on beneficial use, strong governance and safety oversight, and controlled access with enterprise-grade security [4].
Pricing: not yet clear for GPT-6 #
As of October 31 OpenAI's pricing page listed two older Rosalind models (gpt-rosalind-research and gpt-rosalind-discovery) at $5 per million input tokens, $0.50 cached, and $25 per million output tokens, with billing for gpt-rosalind-research effective October 5. There was no separate row for gpt-6-rosalind. The help centre says only that GPT-6 Rosalind Research has different standard rates from the Rosalind 5.5 models [5]. API reseller APIYI reports $15 input and $75 output per million tokens — 1.5 times Astra's $10 and $50 — but that figure does not appear on OpenAI's own pricing page [3].
European considerations: data residency, GDPR, AI Act #
For European life-sciences organisations the buying question shifts from model capability to contractual obligations. OpenAI offers data residency in Europe for eligible API projects and ChatGPT workspaces released on or after March 5, 2026, with a 10% uplift on regional processing. Nothing published says whether gpt-6-rosalind is eligible [3].
Genetic and health data are special categories under GDPR Article 9. Large-scale processing triggers a mandatory data protection impact assessment under Article 35(3)(b). The AI Act's research exemption (Article 2(6)) excludes systems "specifically developed and put into service for the sole purpose of scientific research and development." A general vendor model used in a pharma pipeline is unlikely to fit that narrow wording [3].
The pattern continues #
This is the third trusted-access deployment from OpenAI in 2026, following GPT-5.5-Cyber for vetted security teams in May and Astra for Law in September. In each case vetted customers get a model that is more permissive in a sensitive domain, under monitoring. With Rosalind, OpenAI states openly that the model itself is unchanged and only the policy and the gate differ [3].
For buyers, the real assets are the approval, the obligations, and the audit trail — and those sit partly on the customer's side of the contract.
Sources #
1. [OpenAI's New Biology Model Has a Gate Before the Lab - Medium](https://medium.com/@mehulpratapsingh/openais-new-biology-model-has-a-gate-before-the-lab-c0cba26790f0) (medium.com)
2. [OpenAI Ships GPT-Rosalind-6 Biology Model to Trusted Users](https://aiweekly.co/alerts/openai-ships-gpt-rosalind-6-biology-model-to-trusted-users) (aiweekly.co)
3. [GPT-6 Rosalind shows OpenAI now sells access, not just models, and ...](https://www.i-scoop.eu/gpt-6-rosalind-trusted-access) (i-scoop.eu)
4. [Introducing GPT-Rosalind for life sciences research | OpenAI](https://openai.com/index/introducing-gpt-rosalind) (openai.com)
5. [GPT-Rosalind for life sciences research | OpenAI Help Center](https://help.openai.com/en/articles/20001193-gpt-rosalind-for-life-sciences-research) (help.openai.com)