Theo & Ben break down OpenAI's latest model, Astra, and why it's their new benchmark for AI coding, computer use, and multimodal work. But it's not all good: they explain why its UI, stopping behavior, and agent reliability still create friction in real software workflows. From DEFCON puzzles to coding-agent PRs, we're comparing Astra with Fable and asking what a trustworthy OpenAI model should do next.

Thank you to PostHog for sponsoring today's episode!

• PostHog, all-in-one suite of product tools: https://nerdsnipe.link/posthog

• 
  

Listen wherever you get your podcasts:

• Spotify: https://nerdsnipe.link/spotify
• Apple: https://nerdsnipe.link/apple
• Elsewhere: https://nerdsnipe.link/listen

• 
  

Sources available on our Substack:

• https://nerdsnipe.substack.com/

• 
  

Timestamps

00:00 Meet Astra

04:03 Best Model Ever, With Catches

06:33 Reasoning and 3D Benchmarks

20:00 Astra Rebuilds Ping.gg

30:04 Instruction-Following Problems

44:47 The Uncommitted Fix Debate

01:09:37 The PR Babysitting Failure

01:22:41 Why Fable Still Wins

01:30:41 Multimodal and Computer Use

01:39:29 Astra vs. Fable Fleet Data