OpenAI GPT-6.2 Astra is Here – It’s Not What You Think

OpenAI GPT-6.2 Astra is Here – It’s Not What You Think

More

Summary

OpenAI has released two models, GPT-6.2 Astra and GPT-6.2 Soul, and Matt Wolfe breaks down what is new. Astra is the flagship for demanding projects, comparable to Anthropic’s Fable, while Soul is a cheaper everyday model similar to an Opus tier. Astra costs $8 per million input tokens and $40 per million output tokens, 20 percent less than the previous version and the same as Claude Fable 5.2. Soul stays at $2 and $10. Both are available in ChatGPT, Codex and the API on paid plans.

The pitch is efficiency rather than raw intelligence. On Artificial Analysis, Fable 5.2 still leads at 65 while Astra scores 62, ahead of Gemini 4 Carbon and Claude Opus 5.5, at about $4 per task versus roughly $26. Coding scores did not move. Wolfe summarizes mixed early feedback from Theo, Matt Shumer, Dan Shipper, Ethan Mollick and others, who describe the model as hard to work with even if not unintelligent. He also shares his own test, a Mega Man-style game clone, with generally positive results.

The video closes by explaining OpenAI’s claim that Astra is the first public model to reach its high threshold for AI self-improvement. That means it helps researchers roughly like an extra research engineer, for example cutting small-model training time by about 43 percent versus 72 percent for the best human result. OpenAI says progress is not yet accelerating beyond its existing trend line.


📺 Source: Matt Wolfe · Published November 18, 2026
🏷️ Format: News Analysis

1 Item

Channels

1 Item

Companies