GEPA basics

Recipe — judge optimization basics. Pointer to the SDK judge_optimization sample.

Canonical sample: samples/core/judge_optimization.py — estimate, run, and apply judge optimizations.

SDK pattern

# Estimate cost first
estimate = client.judge_optimizations.estimate(judge_id=judge.id, budget="medium")

# Start optimization (async)
optimization = client.judge_optimizations.create(judge_id=judge.id, budget="medium")

# Poll until done
while True:
 optimization = client.judge_optimizations.get(optimization.id)
 if optimization.status.value in ("success", "failure"):
 break

# Apply on success — creates a new judge version
if optimization.status.value == "success":
 client.judge_optimizations.apply(optimization.id)

Budget levels: "light", "medium" (default), "heavy". See Bootstrap a judge before GEPA for the week-1 setup.

See also

• Tutorial 5: GEPA optimize
• Concept: Judges