OpenAI’s gpt-6-rosalind costs $75 per million output tokens, above its GPT-6 flagship

A new model called gpt-6-rosalind appeared on OpenAI’s API pricing page, listed at $15.00 per million input tokens and $75.00 per million output tokens. The page says access is limited to approved internal research through a trusted-access program.

The listing sits in a section headed “Life sciences models,” described as OpenAI’s latest Rosalind models. At the time of writing, the page shows a price and an access rule. It gives no description of what the model does, and we found no separate OpenAI announcement for it. This report covers only what the page shows.

What the pricing page shows for gpt-6-rosalind

The page lists three Rosalind models. Prices below are per million tokens.

• gpt-6-rosalind: $15.00 input, $75.00 output, $1.50 cached input. Prompt-cache writes cost $18.75 for a 30-minute window and $37.50 for a 12-hour window.
• gpt-rosalind-research: $5.00 input, $25.00 output, $0.50 cached input. No cache-write prices are shown.
• gpt-rosalind-discovery: $5.00 input, $25.00 output, $0.50 cached input. No cache-write prices are shown.

The page states that access is “limited to approved internal research through the trusted-access program.” It adds that all eligible organizations will keep getting access to the latest GPT-Rosalind models as they are released. Regional processing (data residency) endpoints carry a 10% uplift for eligible models released on or after March 5, 2026.

The page does not say when gpt-6-rosalind becomes available. It gives no context window, no benchmark results, and no list of supported regions. It also does not explain how an organization qualifies for the trusted-access program.

Priced above the flagship it shares a name with

The page itself allows a direct comparison. The GPT-6 flagship tier lists gpt-6-astra at $10.00 input and $50.00 output. At $15.00 and $75.00, gpt-6-rosalind is 50% more expensive on both counts. It also costs more than gpt-6.1-sol ($2.00 and $10.00) and gpt-6.1-terra ($1.00 and $5.00) by a wide margin.

Another specialized line on the same page is priced closer. The cyber model gpt-5.6-cyber is listed at $12.50 input and $75.00 output, so its output price matches Rosalind’s. The two research-oriented siblings, gpt-rosalind-research and gpt-rosalind-discovery, cost a third of the new model’s input price and output price.

The earlier model is a useful reference point. When OpenAI introduced GPT-Rosalind in April 2026, the-decoder reported it as a reasoning model for life sciences research. According to reports of that launch, OpenAI said it beat GPT-5.4 on 6 of 11 LABBench2 tasks, which cover areas such as literature research, database access, sequence manipulation and protocol design. Reports also cite a collaboration with Dyno Therapeutics in which the model ranked above the 95th percentile of human experts on RNA prediction tasks in controlled settings. These are OpenAI’s own figures as relayed by secondary sources, and we have not independently verified them. We found nothing that ties them to gpt-6-rosalind specifically.

Who can use it, and what the page leaves open

For most developers, nothing changes yet. The trusted-access program restricts the models to approved research, so a standard API key will not necessarily be enough. Organizations already in the program are the likely first users, since the page says eligible organizations will keep receiving the latest Rosalind releases.

OpenAI has recently pointed to the family’s research use. On October 21 it shared on social media that GPT-Rosalind surfaced a new lead for drug discovery, asking whether restoring cholesterol transport could protect neurons from degeneration. That is a company claim about a research result, and the post does not tell us which model version produced it.

Several questions remain open. The page does not say whether gpt-6-rosalind replaces the earlier Rosalind models or runs alongside them. It does not say how it differs from the research and discovery variants, or why those two are priced lower. Nor does it say whether the long-context pricing column applies to this model, as the text we could read left that blank.

The pricing page is also the only place the name has surfaced so far. OpenAI has not published a post, system card or documentation page for gpt-6-rosalind that we could find. Until it does, the price and the access limit are the only confirmed facts. Readers should treat any claims about the model’s capabilities as unconfirmed.

For context on how OpenAI is packaging API access elsewhere, see our report on ChatGPT Pro’s monthly API credit.