A cookbook reports that reflective prompt optimization with DSPy GEPA raised accuracy on NuminaMath-1.5 by 11%, for a total cost of less than US$0.50.
A cookbook on DSPy GEPA describes a reflective prompt-optimization approach and gives an example using separate models for inference and reflection. According to the publication, the method increased accuracy on NuminaMath-1.5 by 11%, at a total cost of less than US$0.50.
That result is specific to the reported experiment and does not guarantee the same gain on other models or tasks. If you use AI to study or adapt the material, avoid including personal data or sensitive internal content; use synthetic or anonymized examples and follow your organization’s data rules.