On October 10, TestingCatalog noticed that a model that has not been officially released, gpt-rosalind-discovery, had quietly appeared on OpenAI's API pricing page. It is listed under “Life Sciences Models” and priced exactly the same as gpt-rosalind-research ($5 per million input tokens and $25 for output). But the official documentation page returns a 404, and apart from the price OpenAI has revealed nothing else.
Key points:
- Pointing to drug discovery: Outsiders guess from the name “discovery” that it focuses on screening candidate molecules, split from the research version by task. Different stages demand very different things of a model, so focused optimization on a single stage can be monetized quickly once its capability is proven.
- A vertical commercialization sample: Rosalind has become OpenAI's life-sciences brand (first launched in April, billing since October 5; Amgen, Moderna and others are already connected). It is the first project in which a leading general-purpose vendor gives a specific research field its own brand, its own access controls and continuously iterated models. OpenAI is acting only as a tool supplier to pharmaceutical companies, not entering drug development itself.
- Challenges: The drug discovery pipeline is tightly interlinked, so the more finely the models are sliced, the more handoffs between them are needed. That the two versions are priced identically suggests the differentiated positioning has not yet been settled. Whether it works will depend on later evaluations and experimental evidence.
Source: read the original. This text is a summary for reference only.
