Evaluation experiments have shown that OpenAI's GPT-6 Astra demonstrates high capabilities in code review.
In terms of "executable bug coverage," one of the evaluation metrics, Astra detected approximately 4% more bugs than GPT-5.6 Sol and approximately 22% more than Opus 5.
The difference is particularly pronounced in complex reviews spanning multiple files. Astra recorded a detection rate 20% higher than Sol and 33% higher than Opus 5. This suggests an improved ability to capture the relationships between information dispersed across the entire codebase.
On the other hand, users should be mindful of the usage costs. The API pricing for Astra is reported to be $10 per 1 million input tokens and $50 per 1 million output tokens.
Based on this pricing, calculations for a fixed amount of token usage show that Astra's cost reaches approximately 2.5 times that of Sol, 4.7 times that of Terra, and 47 times that of Luna. OpenAI stated that it is important to evaluate based on the total cost of task success.
Source: GPT-6 Astra in code review: Gains, privacy, and cost (Hacker News Frontpage, 2026-09-05)