GPT-6 Astra in code review: Gains, privacy, and cost

OpenAI's GPT-6 Astra shows modest improvements in code review, catching approximately 4% more bugs than GPT-5.6 Sol overall, with significantly larger gains of 20-33% on complex cross-file reviews where bugs are distributed across the codebase. The model's strength lies in connecting contextual information across files to identify issues that wouldn't be caught by simpler models, though teams must weigh these gains against increased API costs.