GPT-6 Rosalind Research launches with separate approval and higher token rates
OpenAI has begun rolling out GPT-6 Rosalind Research to a limited group of approved life-sciences research customers. The October 30 release reaches ChatGPT Work, Codex and separately approved API organizations, but not the Chat tab. It is a new Research model, not an automatic upgrade for everyone with Rosalind Discovery access.
The practical questions are access and billing: which tier an organization has, whether an administrator has enabled the model, and which rate card applies. OpenAI’s release notes identify the API model as gpt-6-rosalind and say GPT-5.5 Rosalind Research remains available for existing workflows.
Research approval and Discovery approval are different
| Existing access | What the October 30 release means |
|---|---|
| Rosalind Research | GPT-6 Research is rolling out to a limited approved group; GPT-5.5 Research remains selectable. |
| Rosalind Discovery | Remains on 5.5. Discovery approval does not grant either Research model. |
| Enterprise Research workspace | GPT-6 starts in a two-week Admin Preview, off by default. Admins can enable approved users. |
That distinction matters for readers who followed the earlier Rosalind Discovery pricing listing. A Discovery entry was not evidence of access to this newly announced Research model. The two stories concern different access tiers.
What to check if the model is missing
OpenAI’s current setup guide separates approval, activation and user permissions. A subscription, workspace membership, Rosalind Workbench or the Life Sciences Research plugin does not independently grant model access. Existing Research approval also does not bypass Enterprise Admin Preview.
For Work or Codex, check the approved workspace, completed activation and your model permissions with its owner or administrator. For API use, check the separately approved API organization rather than assuming a ChatGPT workspace approval carries over. OpenAI limits API access to approved internal research, not customer-facing products or external commercial applications. Its Academic Plan does not include API access or API credits.
A useful access checklist therefore has separate boxes for the organization, the person and the product. Finding a plugin installed is only one setup observation; it does not settle any of those approval questions. This is an interpretation of the documented access process, not a test of an enabled Rosalind account.
GPT-6 Rosalind Research pricing
The API pricing page lists the following token rates in its life-sciences table. Dollar amounts are per one million tokens, not a monthly subscription or a price per completed research task.
| API model | Input | Cached input | Output |
|---|---|---|---|
| gpt-6-rosalind | $15 | $1.50 | $75 |
| gpt-rosalind-research | $5 | $0.50 | $25 |
| gpt-rosalind-discovery | $5 | $0.50 | $25 |
The displayed GPT-6 rates are three times the corresponding older Rosalind rows. That arithmetic describes the listed token prices, not research quality, speed or cost per successful result. The API table also lists GPT-6 cache-write rates of $18.75 for 30-minute retention and $37.50 for 12-hour retention, per million tokens.
Work and Codex credits are a different unit
The credit-based rate card lists Standard Work and Codex usage separately from API dollars:
| Work / Codex model | Input credits | Cached-input credits | Output credits |
|---|---|---|---|
| GPT-6 Rosalind Research | 375 | 37.5 | 1,875 |
| GPT-5.5 Rosalind Research | 125 | 12.5 | 625 |
| GPT-5.5 Rosalind Discovery | 125 | 12.5 | 625 |
For a purely illustrative 100,000 uncached input tokens and 20,000 output tokens, GPT-6’s listed Standard rates produce 75 credits: 37.5 for input plus 37.5 for output. The same token quantities at the 5.5 Research rates produce 25 credits. This example does not predict how many tokens either model needs for a real task. Codex does not charge for cache writes.
A dollar example, with the limits kept visible
Eligible Enterprise token-based agreements have their own USD rate card. Its Standard Work and Codex rates for GPT-6 Research are $15 input, $1.50 cached input and $75 output per million tokens. The corresponding 5.5 Research rates are $5, $0.50 and $25. The customer’s agreement determines which billing model applies.
Using the same hypothetical 100,000 uncached input and 20,000 output tokens gives a $3 model-token charge for GPT-6 Research versus $1 for 5.5 Research. Additional applicable feature charges, contractual discounts and the actual tokens consumed can change a bill. These are calculations from published rates, not ByteForward benchmark results or a total project estimate.
For a team deciding what to do next, access verification comes before migration planning. Once authorized, compare actual workload consumption and useful output rather than treating a model-generation number or token-price ratio as a performance result. ByteForward has not tested this model, and this article makes no claim about biological outcomes or clinical suitability.
Featured image: AI-generated editorial illustration by ByteForward, with OpenAI’s official wordmark. Not a product interface or photograph of a research facility. Illustration: ByteForward. Wordmark: OpenAI.
