GEPA Shows Significant Gains After Retest

strickvl · x · 2026-07-13

The author corrected their previously reported GEPA improvement data, realizing the original comparison had two flaws:

They then ran a new experiment: same model, same reward, using the new GEPA system prompt, and tested on samples GEPA had never seen before. The results showed:

They concluded that the results indeed seem to contain a real signal, while reminding everyone to fully understand how metrics are derived to avoid having to issue retroactive corrections.

Related event: Prime Intellect Releases v1 API, GEPA Data Corrected(2 posts)→

Original post →

More from coding & agent

coding & agent channel →