The listing is the news

OpenAI has not announced GPT-Rosalind Discovery. It has not explained what “discovery” changes, published an evaluation, or said who can use it. But its API pricing page listed a model ID — \`gpt-rosalind-discovery\` — under life-sciences models, next to \`gpt-rosalind-research\`.

The listed rates were identical: $5 per million input tokens, $0.50 per million cached input tokens, and $25 per million output tokens. TestingCatalog first reported the entry on October 10 after it was surfaced by an automated model tracker.

That is the full confirmed story. A line on a pricing page proves that OpenAI prepared a billable product name. It does not prove the model is generally available, finished, or meaningfully different from the existing research endpoint.

“Discovery” is still a guess

The label points toward drug discovery, but the existing GPT-Rosalind product already covers early discovery work. OpenAI describes Rosalind as a life-sciences model for biology, drug discovery, and translational medicine. Its help material already lists target discovery, target validation, literature synthesis, genomics, and experiment planning among the supported workflows.

So the new suffix could mean several things. It may be a narrower model tuned for the front end of the drug pipeline. It may be a distinct API configuration for a workflow OpenAI already sells. It may even be an internal naming change that reached the public price table too early. Until OpenAI publishes documentation, choosing among those explanations is speculation.

The price tells us less than it seems

Identical pricing does not mean identical capability. Providers often put different variants on the same rate card when they share an underlying model family or cost envelope. Here, the price gives buyers one practical signal: if Discovery ships at the listed rate, OpenAI is not positioning it as a cheaper screening model or a premium tier above Rosalind Research.

The access model matters more. GPT-Rosalind moved out of research preview in September and is offered globally only to eligible organizations through a trusted-access program. OpenAI says approved organizations receive the latest Rosalind models as they are released. That makes a quiet addition plausible, but it does not establish that current customers can call the new endpoint today.

What to wait for

Do not redesign a research workflow around the model ID yet. The missing pieces are the ones that determine whether a specialist model is useful: supported tools and datasets, versioning, evaluation results, access requirements, safety controls, and whether outputs are reproducible enough for regulated work.

A proper announcement should also settle the basic product question: is Discovery a model, a mode, or a routing label? Until then, this is a credible preview hiding in a price table, not a release.

Sources

  1. [1] TestingCatalog — OpenAI lists new gpt-rosalind-discovery model in docs (Oct 10, 2026)Read source
  2. [2] OpenAI — API pricingRead source
  3. [3] OpenAI — Introducing GPT-Rosalind for life sciences researchRead source
  4. [4] OpenAI Help Center — GPT-Rosalind for life sciences researchRead source