OpenAI appears to have leaked its next life-sciences model. On October 10, AI tracker accounts spotted "gpt-rosalind-discovery" listed on OpenAI's API pricing page under a "Life sciences models" section, sitting right beside gpt-rosalind-research. There is no blog post, no help article, no system card — just the listing.
The context matters. GPT-Rosalind was introduced in April as a frontier reasoning model for biology, drug discovery, and translational medicine, with Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific named as early partners. It left research preview on September 11, moved to OpenAI's trusted-access program for eligible organizations, and published pricing took effect on October 5. The pricing page also notes that all eligible organizations will keep getting access to the latest Rosalind models as they're released — this was always meant to be a family, not a single model.
The most telling detail is the price tag. The new model is priced identically to gpt-rosalind-research: $5 per million input tokens, $0.50 per million cached input tokens, and $25 per million output tokens. A "discovery" variant at the same price band suggests either a sibling model tuned for early-stage work like target identification and molecular design, or a differently tuned version of the research model aimed at a narrower job. OpenAI has a habit of listing models on the pricing page shortly before announcing them, so the announcement probably is not far off.
For anyone building with AI, the signal is in the direction of travel. The labs are converging on life sciences as the next serious frontier. The current GPT-Rosalind-5.5 is trained on top of GPT-5.5 and, per its system card, deliberately does not refuse sophisticated biology questions — gated access is the safeguard instead of refusals. Anthropic is working the same territory with Claude Operon, a biology-focused mode in the Claude desktop app, and Claude Science, currently in beta, while OpenAI also runs a ChatGPT for Academic Researchers project. A discovery-focused model would complete a research-lab stack: hypothesis generation and literature synthesis on one side, target identification and molecular design on the other.
The honest caveat: everything about the purpose is still a guess. The name says discovery; the listing says nothing else. Treat this as a watch item, not a launch. But the pattern is familiar, and when the announcement lands, the questions worth asking are whether discovery is a separately trained model or a tuning of the research model, and whether trusted-access remains the gate — because the pricing page answers neither.
