Breaking

Perplexity Uses GPT-6 Astra in Codex to Test End-to-End Systems

September 14, 2026 at 13:05 EDT

Software Dev & Coding
AI Agents
Foundation Models

Perplexity is using GPT-6 Astra in Codex to build test harnesses, simulate third-party service responses, and validate application workflows from end to end, extending AI coding agents from code drafting into operational checks. The company says the model is also being used to write communications, change software, and monitor production systems with fewer human check-ins than earlier models.

September 14, 2026 · Perplexity

Astra Moves Testing End to End

Perplexity is using GPT-6 Astra in Codex to build test harnesses, mock outside services, and check full application workflows with fewer human check-ins.

Coding benchmark context

$10 / $50

API price per 1M input and output tokens

89%

fewer unintended outcomes than GPT-5.6 Sol in an internal computer-use safety benchmark

1.9x

faster task completion with the updated Codex harness on Mind2Web

Perplexity testing loop

Build harness

Create a small program around the app

→

Mock services

Generate realistic API or connector replies

→

Run workflow

Check the application from start to finish

→

Review result

Humans inspect a smaller surface

What improves

End-to-end checks can be assembled faster.
Mocked services reduce reliance on live third-party systems.
Cross-file reasoning helps expose interface drift and authorization bugs.

What still matters

Completion criteria must be explicit.
Approval policies still gate consequential actions.
Website and app restrictions can narrow agent access.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.