OpenAI vets institution, lead researcher and planned use before granting eligible labs up to 30 free GPT-Rosalind seats

The unit of the offer is a seat. Each eligible lab gets up to 30 on the Business Standard plan for up to twelve months ^[1]. That caps the subsidy at 360 seat-months per lab ^[12].

The gate has more than one step, and the two reports describe it differently. The Decoder says OpenAI vets the institution, the lab's lead researcher and the planned use before it grants access ^[3]. RuntimeWire says labs must separately clear OpenAI's trusted-access review to use GPT-Rosalind ^[5]. Either way, the application is a disclosure. The lab states what it intends to do.

The disclosure stays in play after approval. OpenAI monitors how the model is used once a lab is in ^[4]. According to OpenAI, safety checks for registered, highly sensitive applications factor in the approved research context, and the company follows up with researchers when something is unclear ^[6]. We think that makes the declared plan the baseline for later activity. A lab that drifts from it should expect a question from OpenAI.

The missing API fits the monitoring. In our view, activity inside a seat is easier to observe and attribute than calls from a lab's own code.

RuntimeWire reported that the plan could bring the model into academic workflows, while seat limits, approval and the lack of API restrict how labs can use it ^[11]. We think a free year gets the model into labs' routines, and the vetting, monitoring and API exclusion limit OpenAI's risk while it does. The reports do not give the price after month twelve or OpenAI's stated aim for the plan, and the price decides whether a lab renews.