GPT-Rosalind reaches general availability through trusted access

OpenAI's 8 September changelog records general availability for GPT-Rosalind, using the API identifier gpt-rosalind-research. Availability remains subject to the trusted-access programme, rather than open access for every developer.

The Help Centre describes eligible enterprise life sciences research teams using the model in ChatGPT Enterprise, Codex and the API. Approval is required. API access is intended for approved internal research workflows and is not available for customer-facing products or external commercial applications.

OpenAI describes support for biological reasoning, literature synthesis and scientific tool workflows, including early discovery and hypothesis generation. The published standard token rates are $5 input, $0.50 cached input and $25 output per million tokens. Billing begins on 5 October 2026; these are scheduled rates, not a statement that billing had already started when this page was prepared.

Begin with an access and data-governance review before designing an integration. Assign an owner for the scientific question, list the permitted data sources and define what evidence a researcher needs to inspect before accepting an answer. Model output should be treated as a proposed interpretation that requires domain review.

Keep source references, tool versions and analysis parameters with the result so another researcher can reproduce the workflow. Separate a generated hypothesis from experimental evidence, and record uncertainty explicitly. For a pilot, choose a narrow literature or data-analysis task with an existing human baseline rather than attempting to automate an entire discovery programme.

Approval for a workspace should not be assumed to grant every API or product use. Verify the specific surface and workflow with OpenAI before implementation. These are AZ Labs planning recommendations, not a claim that we have access to or have evaluated GPT-Rosalind.