Users are reporting a decline in the performance of OpenAI's GPT-6 Astra model just one week after its launch. Developers on social media platforms claim the model provides faster but lower-quality responses compared to its initial release state.
Several users conducted side-by-side comparisons using identical prompts from the launch day and today, noting a visible drop in output quality. Some developers, including the team behind Opencode, have reverted to the previous model, GPT-5.6 Sol, citing higher costs and performance issues.
Speculation among users suggests that OpenAI may have adjusted the model's computing effort or applied quantization to reduce operational costs. Similar concerns regarding performance degradation were previously raised about the GPT-5.6 Sol model in July.
Despite the criticism, some observers argue that the model's performance has remained consistent, suggesting that initial user enthusiasm masked its limitations. Others note that the model exhibits high inconsistency in its reasoning capabilities.
OpenAI has not officially commented on the reports regarding Astra. The model is currently priced at $10 per million input tokens and $50 per million output tokens, which is 2.5 times the launch price of the Sol model.
