Tools
OpenAI’s GPT-6 Rosalind Is a Biology Research Model Almost Nobody Can Use Yet
OpenAI has released GPT-6 Rosalind Research, its new life-sciences model for biology, drug discovery and translational medicine, but almost nobody can use it. Access is limited to approved research organizations under a trusted-access review. It is not in the ordinary ChatGPT chat window, and buying a subscription alone does not provide access.
What it is
According to OpenAI’s October 30 safety report, the model uses the same underlying weights as GPT-6 Astra, without additional training for this release. It has fewer refusals on dual-use biological questions that also have legitimate research uses, while still refusing requests for biological weaponization. OpenAI classifies the launch as High capability in the Biological and Chemical domain under its own risk framework, below its critical threshold, and says none of its four Critical biological capability evaluations reached the relevant threshold.
The benchmark numbers, and their limits
At the highest reasoning budget tested, the newer model scored 49.6% on a medicinal-chemistry evaluation, compared with 28.9% for its predecessor GPT-Rosalind-5.5. On a genomics and quantitative biology benchmark it scored 37.7%, and on an applied life-sciences benchmark built from tasks contributed by working scientists it averaged 61.3%, versus 44.0% for the predecessor.
Those are OpenAI’s own results on its own tests, and they need two caveats. One biological-data benchmark used a configuration different from the production system, OpenAI notes. And benchmark scores do not establish that the model can safely diagnose patients, choose treatments, or deliver a successful new medicine. They show research-task performance, not clinical value.
Who gets in, and what it costs
Approved Research organizations get the model in the model picker in ChatGPT Work and Codex, where it pairs with the Life Sciences Research plugin, but not in the Chat tab. API access, under the model name gpt-6-rosalind, requires a separate approval and is limited to internal research tools, not customer-facing products. Applicants must pass a review covering their proposed research, security controls and governance, and approved customers must restrict use to authorized users and apply qualified human review to outputs. Approval for the Discovery tier does not grant access to the Research model.
In Enterprise workspaces, the model begins with a two-week administrator preview and is off by default during that period. Pricing is $15 per million input tokens, $1.50 for cached input, and $75 per million output tokens, three times the price of GPT-5.5 Rosalind Research at $5, $0.50 and $25. For the labs that qualify, that is the cost of a frontier biology specialist. For everyone else, Rosalind is a model to watch, not a tool to try.