GPT-5.6 Luna versus GPT-6 Astra

The GPT-5.6 Luna model detected 69 confirmed errors for $0.20, while the GPT-6 Astra model found 92 errors for $5.66.

This means that the Luna model is 28 times cheaper, but 24% less accurate, especially when it comes to detecting security errors.

These results sparked a discussion on news.ycombinator.com about whether a $1.20 model is good enough for code review.

It's worth noting that the GPT-5.6 Luna model showed good results despite its low cost.

This comparison may change the approach to choosing models for code review, as developers can now consider not only accuracy but also the cost of the model.