# Developers Criticize GPT-6.2 Astra on Long Tasks and Creative Work

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/62228
Published: 2026-11-20T17:36:29.000Z
Updated: 2026-11-20T17:36:29.000Z
Section: Technology

> Matt Shumer criticized Astra’s agent-workflow performance, while Theo Browne described stalled project work and similar game-level designs.

Developers and frequent AI users have criticized GPT-6.2 Astra’s performance on long-running tasks, agent workflows and creative work, raising questions about how reliably it handles extended projects.

Matt Shumer, an AI investor and former HyperWrite CEO, said after testing GPT-6.2 Astra and GPT-6.2 Sol that both models were nearly unusable and that he did not recommend them. In an agent workflow, Shumer said, the models supervising Astra had to redo work assigned to it.

T3 Code founder Theo Browne said he would not choose Astra over Claude Opus 5.5 for his daily development work. He described cases in which Astra’s explanations left users unsure what it had completed, where it was stuck and what they should do next.

Browne said Astra spent about two hours on a task to design a new level for Outer Wilds before asking where to write internal files and video output. In a separate example, he said a seven-day effort to improve a project produced no substantive improvement and slowed compiler performance by 56%. Another four days brought no progress, Browne said.

Browne also questioned Astra’s creative output. He said he and another tester asked for original levels for different games, but the designs both featured water and bells and shared the name “Bellwether.” Some testers praised Astra’s work in Three.js, Blender and 3D visualization.

OpenAI released GPT-6.2 Astra alongside GPT-6.2 Sol on Nov. 17. Astra is aimed at demanding reasoning and coding work, while Sol targets everyday coding and agent workflows. OpenAI also cut Astra’s standard API input and output prices by 20% at launch.

The criticism follows earlier coverage of [Astra’s coding evaluation results](<https://www.tokenpost.com/news/technology/61043>) and [its High rating for AI self-improvement](<https://www.tokenpost.com/news/technology/59763>).



## Links in this article

- [Astra’s coding evaluation results](https://www.tokenpost.com/news/technology/61043)
- [its High rating for AI self-improvement](https://www.tokenpost.com/news/technology/59763)
