OpenAI Pricing Page Lists Unannounced gpt-rosalind-discovery
The OpenAI API pricing page has listed gpt-rosalind-discovery, priced the same as gpt-rosalind-research, at $5 per million tokens for input and $25 for output. OpenAI has not officially announced this model.
This series is only available to institutions approved through the trusted access program for internal life sciences research, with a caching input fee of $0.5 per million tokens and no charge for cache writes. Qualified institutions will continue to receive access to the latest GPT-Rosalind models.
The research version began billing on October 5. The discovery version is priced the same as research, with no specific differences stated by the official source, but the series is positioned for life sciences, supporting early discovery workflows.
GPT-Rosalind was launched on April 16 as a research preview, named after Rosalind Franklin, optimizing biological reasoning, protein engineering, genomics, and multi-step research processes, with the Codex life sciences plugin connecting to over 50 databases and tools.
Early partners include Amgen, Moderna, Thermo Fisher Scientific, and Allen Institute. The model scored 0.751 on the BixBench benchmark and surpassed general models on multiple tasks in LABBench2.
Approved biopharmaceutical institutions can purchase dedicated model usage rights through the API and enterprise plans, contributing to OpenAI's vertical model revenue. Ordinary users cannot access it, while pharmaceutical companies benefit from accelerated early discovery and hypothesis generation. OpenAI strengthens its pricing power and customer lock-in in high-value industries through access barriers.
The pricing is close to some flagship models like gpt-5.5, but with stricter limitations, restricted to internal research use and not supporting customer-facing products or external commercial applications.
Source: Public Information
ABAB AI Insight
OpenAI launched GPT-Rosalind in April 2026 as the first domain-specific cutting-edge reasoning model, initially providing access to Amgen, Moderna, Thermo Fisher, etc., in a research preview, while also releasing a free Codex plugin connecting to multi-omics databases, named in tribute to Rosalind Franklin's contributions to DNA structure.
The company filters institutions through the trusted access program, embedding the dedicated model into corporate internal research and Codex workflows, concentrating resources on high-value biopharmaceutical clients. The motivation is to capture pharmaceutical R&D budgets and establish compliance barriers, similar to its approach in expanding Trusted Access in the cybersecurity field.
This is akin to Google DeepMind focusing on protein structure prediction with AlphaFold and collaborating with pharmaceutical companies, or Anthropic's controlled deployment in specific high-risk areas. It is currently in a phase of expanding from general models to vertical industry control, prioritizing approved enterprises over the open mass market.
Essentially, this represents a transfer of pricing power and technological substitution: dedicated models transform general capabilities into industry-specific tools through access barriers and domain optimization. The mechanism is that high-risk research requires compliance review and identity verification, thereby binding capabilities to clients, increasing switching costs, and concentrating high-value revenue.
ABAB News · Cognitive Laws
General models sell capabilities, dedicated models sell access
High-value industries lock the door first, then sell the key
Pricing power = vertical depth × access barrier × switching cost