New Model gpt-rosalind-discovery Appears on OpenAI Life Sciences API
nashnova research
A new model name, gpt-rosalind-discovery, has appeared on OpenAI's API pricing page alongside the existing gpt-rosalind-research — with no official announcement and no stated purpose. This means a new product outline is visible, but what it actually does remains blank.
How was the new model spotted?
Tech-tracking outlet TestingCatalog reported on October 10, 2026 that OpenAI's API pricing page now lists gpt-rosalind-discovery.
The model has received no official announcement; the pricing page includes no description of its intended use.
In plain terms = OpenAI did not launch it publicly — an outside observer found it in the price table. Right now it is a name with no feature description.
Same pricing — does that mean the two models are the same?
Both models share identical published pricing: $5.00 per million input tokens, $0.50 cached input, $25.00 output.
This means → the billing structure matches, but that does not prove the two share the same training, tool access, or task performance.
In plain terms = identical pricing only tells you OpenAI charges the same rate for both. What each model is good at — or whether they differ at all under the hood — the price alone cannot answer.
What has the Rosalind product line done so far?
GPT-Rosalind launched on April 16, 2026, officially covering biology, drug discovery, and translational medicine, with launch-day plugins connecting over 50 scientific tools and data sources.
Partners include Amgen and Moderna. A June 3 capability update emphasized medicinal chemistry and genomics; on September 11 the model exited research preview and opened to qualifying organizations globally.
In published benchmarks, the June update scored 27.5% on MedChemBench — a medicinal-chemistry benchmark — versus 25.1% for GPT-5.5, a gap of 2.4 percentage points. These results belong to the existing Research version and do not apply to the Discovery model, which has no independent evaluation.
Is "Discovery" specifically for drug discovery?
TestingCatalog suggested Discovery may be aimed at drug discovery, but acknowledged that its purpose is unclear.
This means → the name offers a directional clue — "discovery" in pharma typically refers to the early stage of finding new drug candidates — but a name cannot substitute for an official task list and capability description.
Whether Discovery is a standalone model, a fine-tuned variant, or another product arrangement remains unconfirmed. The pricing-page footnote names gpt-rosalind-research when describing the October 5 billing date; Discovery's activation schedule and access terms are not addressed.
What would it take to judge this new model's value?
According to BioTender's analysis, evaluating the new model requires answering at least three questions: what tasks it optimizes for, its concrete gains over Research on the same benchmarks, and whether experimental results back those gains.
In plain terms = until an official disclosure arrives, Discovery is a confirmed model name plus a set of unverified product hypotheses.
This reflects a broader pattern: AI productization in life sciences is iterating so fast that new model names now appear faster than official explanations can follow.
市场有风险，内容仅供研究参考，不构成投资建议。