OpenAI releases GPT-Rosalind-6 biology model to approved research users only
On October 30, 2026, OpenAI published a system card addendum for GPT-Rosalind-6, a reasoning model for biology, drug discovery and translational medicine. The model runs on the same weights as GPT-6 Astra, the company's flagship, and OpenAI says it added no training data and no extra post-training for this release. It is offered as GPT-Rosalind-6 Research to qualified scientists, research institutes, companies and government partners.
Access runs through a review rather than open sign-up. OpenAI checks whether an applicant has a life sciences mission, a credible use case, sound research governance, and the ability to control access. Approved organizations must restrict use to authorized people with a legitimate need, apply least-privilege and insider-risk controls, and require qualified human review before outputs shape experiments or decisions. According to Forkast, the model is used in Work and Codex, API use needs separate approval, and it is not available in the Chat tab. A two-week Admin Preview is off by default.
On safety, the model refuses fewer dual-use biology queries than GPT-6 Astra but still refuses requests that would meaningfully enable biological weaponization. OpenAI rates it High for biological and chemical capability, below its Critical threshold. At the highest reasoning budget, its medicinal chemistry score was 49.6% against 28.9% for GPT-Rosalind-5.5, and genomics single-attempt success reached 37.7% against 16.7%. Biosecurity red-teaming was run on GPT-Rosalind-5.5 before launch and is being repeated for the new model.
Why it matters
For founders in life sciences, the release sets a template: a general tier for everyone and a vetted tier for dual-use research. A team that wants the higher-capability model has to clear an access review and show security and governance controls before it can build on it. Governance paperwork, identity and access controls, and human review steps belong in the product plan from the first sprint, not at procurement. Customer-facing products are a separate question, since approval for internal research tools does not allow the same model to be exposed to outside users, so anyone selling into research customers should get a written answer from OpenAI first.
For investors, a gated model changes what a life sciences AI company can count on. A research workflow that depends on the model stops if approval is lost, so access continuity belongs in diligence. Reports put API pricing at $15 per million input tokens and $75 per million output tokens, which buyers will weigh against the review burden and the approval requirements when judging revenue potential.
Key points
- 1OpenAI published the GPT-Rosalind-6 system card addendum on October 30, 2026.
- 2The model uses GPT-6 Astra's weights, and OpenAI says no new training data or extra post-training was added.
- 3Medicinal chemistry scores reached 49.6% versus 28.9% for GPT-Rosalind-5.5 at the highest reasoning budget.
- 4OpenAI rates the launch High for biological and chemical risk, below its Critical threshold.
