- OpenAI has expanded research preview access to its updated GPT-Rosalind for eligible life sciences organizations, including Novo Nordisk.
- Novo Nordisk is leveraging GPT-Rosalind to assist researchers in analyzing complex datasets, identifying patterns, and testing hypotheses more quickly.
- "Life sciences research is complex, data-rich, and interdisciplinary. To deliver meaningful value for researchers, advanced AI models must be grounded in trusted scientific data, connected to validated tools, and integrated into the real-world workflows researchers use every day."
- OpenAI evaluated GPT-Rosalind using LifeSciBench, an expert-judged benchmark across six life sciences workflow areas, and reports that GPT-Rosalind outperformed GPT-5.5.
- On MedChemBench, which tests medicinal chemistry tasks, GPT-Rosalind scored 27.5 percent compared to 25.1 percent for GPT-5.5.
- For long-horizon genomics, GPT-Rosalind attained 21.6 percent accuracy on GeneBench, outperforming GPT-5.5, which scored 20.4 percent while using 31 percent fewer tokens.
- OpenAI also introduced LabWorkBench, where GPT-Rosalind achieved 63.2 percent accuracy, exceeding the 55.8 percent score of GPT-5.5.
- Novo Nordisk expressed satisfaction in the partnership with OpenAI, stating their anticipation for exploring how GPT-Rosalind can support drug discovery efforts.
OpenAI has updated GPT-Rosalind and expanded research preview access for eligible life sciences organizations such as Novo Nordisk.1
This development enables researchers to analyze complex datasets, identify patterns, and test hypotheses with greater speed and accuracy.
OpenAI claims that GPT-Rosalind outperformed GPT-5.5 on LifeSciBench, a benchmark that assesses six critical areas in life sciences workflows, including evidence handling and scientific reasoning.4
Specifically, GPT-Rosalind achieved better results on specialized evaluations, showing improvements in various categories.
On MedChemBench, which evaluates medicinal chemistry tasks, GPT-Rosalind scored 27.5 percent compared to 25.1 percent for GPT-5.5.5
In tests related to genomics, GPT-Rosalind scored 21.6 percent in accuracy on GeneBench while utilizing 31 percent fewer tokens than its predecessor.6
Additionally, OpenAI introduced LabWorkBench to support real wet lab protocols, where GPT-Rosalind attained 63.2 percent, compared to 55.8 percent for GPT-5.5.7
This significant enhancement marks a pivotal moment in how advanced AI models can assist in life sciences research, making the integration of AI into research workflows increasingly viable and effective.
“OpenAI has updated GPT-Rosalind and granted access to Novo Nordisk, facilitating data analysis in complex life sciences research. The partnership aims to improve drug discovery through advanced AI-driven insights.”
EdTech Innovation Hub