OpenAI: Perplexity Uses GPT‑6 Astra for Testing and Production Monitoring, Study Says
OpenAI says in a customer study that Perplexity uses GPT‑6 Astra for communications, system changes, workflow testing, and production software monitoring. The model’s permissions were not disclosed.
Perplexity GPT‑6 Astra is used for testing, software changes, and monitoring production systems, according to an OpenAI customer case study. OpenAI published the study on September 14, 2026. However, the publicly available material does not specify what permissions the model receives in the production environment or whether it can make changes without human approval.
In the study, OpenAI quotes Johnny Ho, co-founder and Chief Strategy Officer of Perplexity. According to his statement, the company uses GPT‑6 Astra to create communications, modify live systems, and monitor production software.
Perplexity GPT‑6 Astra in Testing Workflows
According to the quote, the model also helps create small test programs. These simulate responses from external services and are used to verify the entire workflow. This is a type of testing that checks not only an isolated part of the code, but also how multiple steps work together, including communication with external interfaces.
Ho also stated in the OpenAI material that, for similar tasks, the team checks the model less often than older generations. This claim is attributed to a Perplexity representative in the model provider’s reference study, not to an independent measurement.
The description suggests that AI is being used beyond assistance with writing code or text: GPT‑6 Astra is intended to be part of testing and operational workflows. However, the available information does not indicate that Perplexity has given the model fully autonomous control of the production environment.
Details on Permissions and Controls Are Missing
The study does not state the exact scope of GPT‑6 Astra’s production permissions. It is therefore not known whether the model only proposes changes, prepares them for review, carries them out in a restricted environment, or can deploy them directly. Information about approval processes, access restrictions, or other protective mechanisms was also not disclosed.
The material also contains no measurable data on reliability, error rates, the number of incidents, or the impact of deployment on the Perplexity product. The study therefore does not provide enough data to compare the stated approach with other ways of using models in software operations.
The case study comes directly from OpenAI, the provider of GPT‑6 Astra. The available materials contain no separate technical statement from Perplexity that would independently elaborate on or confirm the scope of the deployment.
What Will Be Important to Watch
It will be important to see whether Perplexity publishes its own technical details about how the model is used, particularly regarding human oversight and approval of changes. Specific metrics on reliability, security controls, and testing results from OpenAI or Perplexity would also be useful.
A separate question is whether GPT‑6 Astra makes changes in production or primarily prepares proposals and tests them under supervision. OpenAI’s customer study does not publicly define this boundary.
Sources
- OpenAI – Perplexity trusts GPT‑6 Astra with end-to-end systems – Confirms publication of the case study on September 14, 2026, and the usage claims attributed to a Perplexity representative.
Verified and updated: 09/14/2026 06:24
