OpenAI has released GPT-Rosalind-6, a life-sciences version of GPT-6 Astra that uses the same weights, and a new gpt-6-rosalind row on its API pricing page lists $15.00 per million input tokens and $75.00 per million output tokens. GPT-6 Astra is listed at $10.00 and $50.00, so the same weights cost 1.5 times as much on both, by our arithmetic.
The system card addendum, published on 30 October, says: “We did not add training data or perform additional post-training for this release.” What changes is the access and refusal policy. OpenAI writes that “Rosalind has fewer refusals on dual-use biological queries but is only made available via trusted access”.
Short explainer · Rosalind-6 at 1.5x Astra
A vertical news Short summarising this story. Figures come from OpenAI’s system card addendum and API pricing page; our arithmetic is labelled.
Business Pill · The Model and the Harness
A short Business Engineer explainer on the difference between an AI model and the harness of tools and policy around it. It teaches the general idea only and does not discuss GPT-Rosalind-6.
The key insight:OpenAI’s system card says GPT-Rosalind-6 uses GPT-6 Astra’s weights with no added training, and its pricing page lists it at $15.00 and $75.00 per million tokens against Astra’s $10.00 and $50.00: 1.5 times, by our arithmetic.
The same pricing page lists gpt-6-astra-law, described as a legal model configuration, at $10.00 input and $50.00 output, which is 1.0 times GPT-6 Astra, by our arithmetic. The page does not use the system card’s name for the model. We read the gpt-6-rosalind row as GPT-Rosalind-6: when we read the same Rosalind table on 11 October, it held only gpt-rosalind-research and gpt-rosalind-discovery, at $5.00 and $25.00 and billed since 5 October, and the new row follows the gpt-6-astra naming.
The page says access to the Rosalind models “is limited to approved internal research through the trusted-access program”. We reported in October that the legal configuration sits at GPT-6 Astra’s own price.
API list prices per million tokens for GPT-6 Astra, its legal configuration, GPT-Rosalind-6 and the gpt-rosalind-research and gpt-rosalind-discovery entries. GPT-Rosalind-6 uses the same weights as GPT-6 Astra, per OpenAI’s system card addendum. Prices from OpenAI’s API pricing page, read 31 October 2026; multiples by our arithmetic; chart by FourWeekMBA.
Who Can Get Access
The addendum says GPT-Rosalind-6 “is available only to approved customers”. Before granting access, OpenAI “reviews whether the applicant has a legitimate life sciences or related mission, a credible use case, appropriate research and operational governance, and the ability to safely control access to the model”.
Non-government applicants also go through business verification and compliance screening, the addendum says. A separate pathway covers “select government entities working on approved biosecurity, biosafety, public-health preparedness, and biodefense use cases”. OpenAI says the release “does not broaden eligibility beyond prior announcements, such as the Rosalind Academic Plan”.
Approved customers carry obligations of their own. They must limit access to authorised users with a legitimate need and “apply qualified human review before model outputs inform experiments, publications, or consequential decisions”, the addendum says.
The Safeguards OpenAI Describes
OpenAI says the model still refuses “requests for biological weaponization” and keeps refusals on other safety topics. Under its Preparedness Framework it treats the launch as High capability in the biological and chemical domain, below Critical, and treats GPT-Rosalind-6 as Critical in cybersecurity, relying on GPT-6 Astra’s assessment.
The addendum says GPT-Rosalind-6 applies “a more conservative cyber refusal boundary” than its base. Replaying known jailbreak attacks with real-time blocking switched off, OpenAI reports a 14.3% task success rate for GPT-Rosalind-6 against 15.7% for GPT-6 Astra, lower being better.
OpenAI says it “can restrict or revoke access for misuse”. It says it tested its monitoring and enforcement through biosecurity red-teaming on GPT-Rosalind-5.5 and is repeating that work for GPT-Rosalind-6, and that it is not publishing the details.
The Business Lever: Access Terms as the Product
OpenAI’s news feed dates the first GPT-Rosalind release to 16 April 2026. Since then it has added a biodefense programme in May, new capabilities in June and an academic plan on 15 October, and we reported the end of its research preview in September and its listing on the API pricing page in October. GPT-Rosalind-6 arrives 197 days after that first release, by our arithmetic.
As we read the two documents, the 1.5x premium is attached to the same model under different access terms and refusal settings, since the system card rules out new training for this release. The pricing page shows the legal configuration at no premium.
The Structural Read
As we read the two documents, OpenAI is pricing the access terms and refusal settings, since the model underneath is the one GPT-6 Astra customers already buy.
As we read it, the vetting described in the addendum is the gate to that premium: mission review, business verification, least-privilege controls and human review before outputs inform experiments.
As we read it, the legal configuration shows the other choice OpenAI has made, since the pricing page lists it at GPT-6 Astra’s own price.
OpenAI, GPT-Rosalind-6 System Card Addendum, 30 Oct 2026
“Rosalind has fewer refusals on dual-use biological queries but is only made available via trusted access.”
Three Implications
FOR LIFE-SCIENCES BUYERS OpenAI says access needs a reviewed mission, use case, governance and access controls, and approved customers must apply qualified human review before outputs inform experiments.
FOR FINANCE TEAMS The pricing page lists GPT-Rosalind-6 at 1.5 times GPT-6 Astra on input and output, by our arithmetic, for the same weights.
FOR OTHER MODEL PROVIDERS The addendum names the access programme and the refusal policy as what changes, and says no training data or post-training was added for this release.
The framework says “It is where decisions are made about who gets access to frontier intelligence, under what conditions, and for which purposes.”
As we read it, GPT-Rosalind-6 puts a price on that layer: the same weights cost 1.5 times as much when sold through a vetted access programme. The framework is the owner’s argument about Anthropic and does not discuss OpenAI, this release or its figures.
What Is Not Established
OpenAI does not say how many organisations have access to GPT-Rosalind-6, and the pricing page does not give a separate figure for government users. The red-teaming of GPT-Rosalind-6 is still under way, by OpenAI’s account. This piece covers access, pricing and safeguards only, and does not report the system card’s biology evaluations.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
GPT-Rosalind-6 runs on GPT-6 Astra’s weights with no added training, yet the pricing row we identify as GPT-Rosalind-6 lists $15.00 and $75.00 per million tokens against Astra’s $10.00 and $50.00, 1.5 times by our arithmetic, and OpenAI sells it only through a vetted trusted-access programme. The difference between the two products is the access and refusal policy that OpenAI wraps around the same model.
A note on sourcing. All facts and figures come from OpenAI’s GPT-Rosalind-6 System Card Addendum (30 October 2026), its API pricing page and its news feed, read on 31 October; multiples and day counts are our arithmetic. We did not contact OpenAI. This piece covers access, pricing and safeguards only.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
