hn.today

I hate GPT-6.2 Astra

twitter.com27 points8 comments
Screenshot of I hate GPT-6.2 Astra

Developer and tech creator Theo argues that GPT-6.2 Astra is not a model he would choose over Opus 5.5 for everyday software work, despite notable strengths. He praises its code quality, ability to catch subtle issues in reviews, computer-use skills, 3D and game development, subagent prompts, and occasional inventive solutions. One such solution helped him design a reverse proxy he says could save $25,000 per month. His central criticism is that these impressive capabilities do not make Astra dependable or pleasant to use on substantial projects.

He identifies three recurring failures: Astra gets stuck asking users for obscure implementation decisions they cannot answer, produces explanations so dense and poorly organized that he often switches to Opus to interpret its work, and can spend days looping without progress. On a TypeScript-to-Rust compiler project, Astra worked for seven days without improvements, made the compiler 56 percent slower while pursuing a size goal, then spent four more days without changes after being redirected toward performance. He also says Astra tends to repeat generic ideas across different tasks. He predicts early excitement over flashy demonstrations will give way to frustration with high usage, opaque code, and stalled work. Although he hopes his assessment is wrong and notes OpenAI addressed some complaints, he considers the model’s long-task reliability and communication serious obstacles.

Read on twitter.com8 comments on Hacker News

Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.

More in AI

The daily digest

Today's best Hacker News stories, summarized and screenshotted, one email a day.