GPT-6.2 Astra state-machine review finds more bugs

quickchat.ai6 min read

The author evaluated OpenAI's GPT-6.2 Astra using QuickReview's state-machine profile on 34 pull requests, comparing it against GPT-6.1 Sol. The new model found 52 of the 55 state bugs in the answer key compared to Sol's 44, but left significantly more non-state comments and cost about 7.6 times as much per pull request. Open article →