OpenAI has put its most capable biology model into the hands of a small, vetted group of drug-discovery teams, and it did so with almost no fanfare. On 30 October the company added a one-line entry to its API changelog announcing GPT-6 Rosalind, a version of its life-sciences model family built for “specialized biological reasoning with scientific tool use”. The same day, OpenAI’s Deployment Safety Hub published a system card addendum that sets out what the model can do, how risky the company considers it, and who is allowed to use it.

For India’s pharmaceutical and biotech sector, which shipped about $16 billion of medicines abroad in the first half of FY27, the release is a preview of how frontier AI for drug discovery will be sold: as a gated, premium research tool, not as another chatbot.

Same brain, different rules

According to the system card, GPT-Rosalind-6 runs on the same weights as GPT-6 Astra, OpenAI’s flagship reasoning model, with no extra training data or post-training. The difference lies in how it behaves and who can reach it. The model refuses fewer legitimate “dual-use” biology questions than the general model, so researchers working on pathogens, protein design or toxicology hit fewer dead ends, but it still declines anything that would meaningfully help someone build a biological weapon.

OpenAI’s own benchmarks show a clear step up over the previous GPT-Rosalind-5.5:

• Medicinal chemistry (MedChemBench v7): 49.6% versus 28.9%
• Genomics and quantitative biology (GeneBench Pro): 37.7% single-attempt success versus 16.7%
• Applied life-sciences research (LifeSciBench Gold): 61.3% versus 44.0%

These are company-run evaluations, and OpenAI says the model is most useful today for early discovery work such as target biology, mechanism studies, literature synthesis and interpreting omics data.

Treated as high-risk by design

Under its Preparedness Framework, OpenAI classifies the launch as High capability in the biological and chemical domain, one notch below Critical. None of the four “Critical” biological-design tests reached its threshold. Because the model inherits GPT-6 Astra’s weights, it is also treated as Critical for cybersecurity and keeps that model’s real-time cyber blocking.

The safeguards are as much contractual as technical. Approved customers must restrict access to named users with a genuine need, apply least-privilege and insider-risk controls, and make sure qualified humans review outputs before they shape experiments, publications or other consequential decisions. OpenAI says it monitors usage and can revoke access if it finds misuse.

Narrow access, premium price

GPT-6 Rosalind sits only in the top “Research” tier of OpenAI’s trusted-access programme and is open to a limited set of approved organisations. It works in ChatGPT’s Work tab and in Codex but not in the regular Chat tab. API access needs separate approval and covers internal research tools only, so customer-facing or commercial products are not allowed. In enterprise workspaces the model starts in a two-week admin preview and is switched off by default.

OpenAI’s pricing page lists the model at $15 per million input tokens and $75 per million output tokens for standard context, according to developer news site sysnote. That is three times the rate for the earlier Rosalind models and more than the general GPT-6 Astra.

The India angle

Since September, OpenAI has offered the GPT-Rosalind family globally to eligible organisations, so Indian drugmakers, CDMOs, CROs and biotech startups can apply through an enterprise or business agreement. Approval is not automatic. Applicants must show legitimate research with public benefit, strong governance and enterprise-grade security.

Indian universities face a bigger gap. OpenAI’s Rosalind Academic Plan, which gives labs up to 30 free ChatGPT Business seats plus Rosalind access, is currently open only to labs at US universities. OpenAI says it plans to expand to other countries but has given no timeline. For now, India’s academic biology labs, many of them closely linked to the country’s generics and vaccine industry, can reach the model only through a paid institutional agreement.

What technology leaders should weigh

• Governance readiness: Trusted access requires named users, biosafety oversight and human review. Pharma CIOs will need these controls in place before they apply.
• Data handling: Indian firms should check where research data is processed and which retention terms apply under their agreement before adding proprietary compound or patient-derived data.
• Cost discipline: At premium token rates, long agentic analyses can become expensive. Teams should test the model on narrow, high-value discovery questions first.

The bigger signal is strategic. Frontier labs are no longer releasing their most powerful scientific capabilities as open, general-purpose tools. They are building gated, audited channels for them. For an industry like India’s, which competes on speed and cost in discovery and development, getting through those gates may soon matter as much as having access to the model.

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!