OpenAI rolls out GPT-6 Rosalind Research through restricted access OpenAI rolls out GPT-6 Rosalind Research through restricted access

The release notes place the model in Work, Codex and the API, outside the Chat tab. Access depends on Rosalind Research approval. An organization with Discovery approval still needs separate Research permission for this rollout. Enterprise administrators can enable approved users during the preview and configure access for the period that follows.

OpenAI describes the release as a change in research access and safeguards around an existing model foundation. Its system-card addendum says Rosalind uses GPT-6 Astra weights, with no additional training data or post-training for this release. The foundation therefore stays the same while the conditions for using it in biological research change.

The access process makes governance part of the deployment work. OpenAI reviews proposed uses, security controls and governance practices before granting access. Approved enterprise users need a legitimate research purpose, and qualified people must review outputs before consequential decisions. A lab planning an integration needs to assign those review responsibilities alongside its technical setup.

Existing GPT-5.5 Rosalind Research workflows remain available under the release notes. That gives approved teams a way to compare their current process with the new deployment. A useful evaluation would keep the research task and human review process fixed, then record whether the new system changes the work required to reach a reproducible result.