Understanding When Reflective Prompt and Skill Optimization Helps on Locally Served Reasoning Models: A Two-Model Study of GEPA and SkillOpt

Key points

• Reflective text-space optimizers such as GEPA and SkillOpt have reported large gains on frontier and mid-size models.1
• For each target, we evaluate the same held-out tasks and four conditions - minimal baseline, GEPA prompt, SkillOpt skill, and manually engineered prompt - on HotpotQA (n=500 per condition) and MATH-500 (n=250).1
• On MATH-500, GEPA raises exact match from 0.828 to 0.904 while reducing latency but increasing token consumption, so lower response time does not imply lower token cost.1
• These results suggest that textual headroom for a model-task pair and the observed effect of a text-space optimizer are more informative when decomposed into content/reasoning, answer-realization/evaluator alignment, and serving-completion/reliability.1

Sources (1)

• [1]Understanding When Reflective Prompt and Skill Optimization Helps on Locally Served Reasoning Models: A Two-Model Study of GEPA and SkillOpt
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 08:45 AM

  We test replication across two reasoning-native target-serving configurations: DeepSeek V4 Flash through DwarfStar and Qwen3.8-27B through LM Studio.

  Reflective text-space optimizers such as GEPA and SkillOpt have reported large gains on frontier and mid-size models.

Extractive summary: sentences quoted from the sources.