Developers Criticize GPT-6.2 Astra Performance
Early users say OpenAI's newly released Astra model struggles with long-running tasks, agent workflows and creative work.
Updated on Nov. 20, 2026 in Artificial Intelligence

Live Poll
Would you pay for an AI tool that struggles with long, multi-step projects?
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Nov. 17, and developers have criticized Astra's performance on extended tasks. Investor Matt Shumer said both models were nearly unusable and that he does not recommend them.
Why it matters
Astra is positioned for demanding reasoning and coding work, so reports of stalled progress and unclear status updates challenge its value for professional developers. The criticism also highlights how difficult it remains to evaluate frontier AI models on real-world, multi-day workflows.
Testers reported that a seven-day project effort produced no substantive improvement and slowed compiler performance by 56%, while OpenAI cut Astra's standard API prices by 20% at launch. Whether these results generalize beyond the cited test cases has not been independently verified.
The players
OpenAI
OpenAI is an artificial intelligence research company that develops the GPT family of large language models, including Astra and Sol.
Matt Shumer
Matt Shumer is an AI investor and former CEO of HyperWrite who publicly criticized the GPT-6.2 models.
Theo Browne
Theo Browne is the founder of T3 Code and a software developer who said he would not use Astra for his daily development work.
The details
Developers including AI investor Matt Shumer and T3 Code founder Theo Browne said Astra struggled in agent workflows, where supervising models had to redo its assigned work. Browne said he would not choose Astra over Claude Opus 5.5 for daily development, while some testers praised its output in Three.js, Blender and 3D visualization.
Timeline
Nov. 17: OpenAI releases GPT-6.2 Astra and GPT-6.2 Sol.
Nov. 20: Criticism of Astra's performance is published.
Tech Race
The criticism places Astra against Anthropic's Claude Opus 5.5, which Theo Browne named as his preferred tool for daily development work. The reports sharpen the competitive contest among frontier AI labs over who delivers reliable performance on extended, real-world coding tasks.
Developers weighing which AI tools to adopt for coding work may find Astra less reliable than alternatives for multi-step tasks. Reported pricing cuts could lower costs for teams that do choose to use it.
The takeaway
Independent testing across varied workloads matters more than launch claims when choosing an AI model for real projects. Users should track how well a model reports its own progress before trusting it with long tasks.
Further reading
Explore more coverage in Artificial Intelligence for context on how frontier models are being evaluated.
Source note: This article includes information reported by TokenPost.
Live Poll
Would you pay for an AI tool that struggles with long, multi-step projects?


