OpenAI releases GPT-6 Rosalind Research for approved life-sciences users
OpenAI's system-card addendum, published 30 October 2026, describes GPT-Rosalind-6. It says the model uses the same weights as GPT-6 Astra, and that OpenAI did not add training data or do further post-training for this release. It says the model is offered as GPT-Rosalind-6 Research to qualified scientists, research institutes, companies and government partners, and only through trusted access. It says OpenAI reviews proposed uses, security controls and governance before granting access. It says Rosalind has fewer refusals on dual-use biological queries, and that it is only made available via trusted access. It says the model is designed to support legitimate biological research, including some dual-use topics, while still refusing requests for biological weaponization and retaining refusals on other safety topics. Under OpenAI's Preparedness Framework, the addendum treats the launch as High in the Biological and Chemical domain, below the threshold for Critical. It treats the model as Critical in cybersecurity and below High in AI self-improvement, using GPT-6 Astra's assessments for those other risks.
Source: OpenAI Deployment Safety Hub (opens in a new tab), 30 Oct 2026Also: OpenAI release notes, 30 Oct 2026 (opens in a new tab)
OpenAI's system-card addendum, published 30 October 2026, describes GPT-Rosalind-6. It says the model uses the same weights as GPT-6 Astra, and that OpenAI did not add training data or do further post-training for this release. It says the model is offered as GPT-Rosalind-6 Research to qualified scientists, research institutes, companies and government partners, and only through trusted access. It says OpenAI reviews proposed uses, security controls and governance before granting access. It says Rosalind has fewer refusals on dual-use biological queries, and that it is only made available via trusted access. It says the model is designed to support legitimate biological research, including some dual-use topics, while still refusing requests for biological weaponization and retaining refusals on other safety topics. Under OpenAI's Preparedness Framework, the addendum treats the launch as High in the Biological and Chemical domain, below the threshold for Critical. It treats the model as Critical in cybersecurity and below High in AI self-improvement, using GPT-6 Astra's assessments for those other risks.
OpenAI's release notes for 30 October 2026 call the same release GPT-6 Rosalind Research. They say it is rolling out to a limited set of approved Business and Enterprise customers in Work and Codex, that access requires Research approval for the organisation and the user, and that it is not available in the Chat tab. They say Enterprise workspaces start with a two-week Admin Preview that is off by default, and that an existing Research approval does not bypass that preview. They say customers can keep using GPT-5.5 Rosalind Research. The API changelog for 30 October 2026 says access is for organisations approved for Rosalind Research, for internal research by authorised users.
What it means for business: this is a trusted-access release, not a general ChatGPT model. OpenAI's own biological and chemical rating for the launch is High, which it says is below Critical. The pages that were opened do not state a token price. This describes those pages. It is not advice on a research use.