On the OpenAI API pricing page, a name that shouldn’t have appeared recently has emerged. A new life science model called gpt-rosalind-discovery, which hasn’t been officially released yet, quietly appeared. Its pricing is exactly the same as that of its older sibling gpt-rosalind-research, which has already been made public. It is assumed that this new model targets the drug discovery process.
This is not a solitary launch. OpenAI has established a brand called Rosalind for life sciences, and has set up multiple product delivery pathways. Previously, it has collaborated with several pharmaceutical companies to explore AI drug development, and has also launched a biodefense program. What’s more notable is its identity: among the leading general-purpose large model manufacturers, OpenAI is the first company to create an independent brand, independent access review, and iterative models specifically for the life sciences field. It maintains a very restrained position for itself, acting merely as a tool supplier—selling specialized models to approved institutions, but never touching the drug pipeline itself, and certainly not planting seedlings in the fields of pharmaceutical companies.
In the field of life sciences, it is highly attractive for large model manufacturers: strong payment capabilities and clear pain points make it a typical high-value-added vertical scenario. OpenAI’s current strategy is to attract leading pharmaceutical companies and research institutions at a low cost, and accumulate data and methods in real-world scenarios. However, reality also presents challenges: the entire AI4Bio field still lacks large-scale clinical validation. Most of the model’s role remains limited to reducing manual labor, and it is far from being able to determine the fate of new drugs independently.
OpenAI is trying to break down models into smaller tasks in life sciences, using this segmented approach to turn investments into profits as quickly as possible. However, the current dilemma is that it’s not clear what the differences are between these two models at the same price, or how much real improvement the new models can bring. Whether it can truly establish itself in laboratories and pharmaceutical companies will depend on evaluation and experimental evidence.