Perplexity: Improving Accuracy with Astra

📅 September 14, 2026

🏷️ 📢 Industry News

⏱️ 60-Second Summary

OpenAI published a customer story on how Perplexity is using GPT-6 Astra to run more of its engineering work. The answer engine uses Astra to write communications, change production software, and watch live systems — and it checks the model’s work far less often than with earlier models. Cofounder and CSO Johnny Ho says every gain in the model’s coding skill also improves Perplexity’s search: better programs for fetching web and internal data, then summarizing them more concisely.

🤔 Why This Matters

Perplexity’s product is accuracy. Stronger code-writing is not a side benefit — it is how the search stack gets faster and more precise. The story also marks a shift from “AI that suggests code” to “AI trusted on end-to-end systems,” with humans stepping in less often.

👨‍💻 Developer Impact

The most useful pattern Ho describes is testing. With little time for manual checks, Perplexity asks Astra to build a small test program around an app. The model then fakes realistic replies from dependent services (an LLM API, a connector, and so on) so the full workflow can be exercised without those services being live. That is the practical takeaway: use the model to generate tests and stand-ins, not only production patches.

🌍 User Impact

Users do not get a new Perplexity feature from this post. They get the claimed downstream effect: better retrieval and shorter, more accurate answers as the code that powers search improves.

🔥 Key Features

• Astra writes comms, edits software, and monitors production with fewer human check-ins.
• Better coding ability is tied directly to better search and summarization.
• End-to-end tests via simulated service responses (APIs / connectors).
• Perplexity says it now trusts Astra with full workflows, not just isolated snippets.

🔗 Official Source

https://openai.com/index/perplexity-improving-accuracy-with-astra