Perplexity moved production to GPT-6 Astra — end-to-end systems under the model's control

Perplexity moved production systems to GPT-6 Astra 🔥

Perplexity moved its production systems to OpenAI's GPT-6 Astra model. Astra covers three things at once: writing communications, changing code, and monitoring production systems.

Why this is a strong signal: this kind of set is usually split across internal tools and a human in the loop. Here, writing communications, software changes, and production monitoring all go through a single model — end-to-end, in production.

What you should look at is the checks, not the fact of the migration. Perplexity verifies Astra's results noticeably less often than it did when working with previous OpenAI models. The frequency of checks is the metric that decides whether an agentic scenario lives in prod or dies in review. You can only cut back on checks where the model consistently produces usable results.

The case is framed as an example of trusting a model in real critical processes, not in a demo. Full breakdown — in the official OpenAI blog