Skip to content
SPECIFICTION
Get listed

GEPA

gepa-ai.github.io

Free plan APISelf-hosted
Visit gepa-ai.github.io
Spec score 8.6 #4 of 28 in AI Prompt Generators

What it is

GEPA is a free, open-source Python package for optimizing prompts, code, configurations, agent architectures, policies, and other text-based artifacts. Its optimize_anything API connects to any system through an evaluator you write, so DSPy or another framework is not required. GEPA examines execution traces and evaluator feedback, uses language-model reflection to diagnose problems and propose targeted mutations, then selects candidates with Pareto-aware criteria. It can optimize API-only GPT, Claude, and Gemini models without model-weight access. Human-readable traces explain changes and support agent debugging. Built-in adapters cover Default, Confidence, DSPy Full Program, Generic RAG, MCP, TerminalBench, AnyMaths, and LangChain, with integrations including MLflow, Comet ML Opik, Pydantic AI, and cookbook projects. You can limit metric calls and set several stopping conditions. GEPA works through API or self-hosted deployment, installs with pip, and uses the MIT license.

Spec sheet

What it does
GEPA is an LLM-based text evolution engine for optimizing prompts, code, configurations, agent architectures, policies, and other text-representable artifacts.Read 4 Oct 2026 · gepa-ai.github.io
Purpose
GEPA is a framework for optimizing prompts, code, agent architectures, configurations, and other text parameters against evaluation metrics.Read 9 Oct 2026 · github.com
Integrations
The project lists integrations with DSPy, MLflow, Comet ML Opik, Pydantic AI, OpenAI Cookbook, Hugging Face Cookbook, and Google ADK.Read 4 Oct 2026 · github.com
Support
The project points users to its documentation, GitHub issues, and a Discord community.Read 9 Oct 2026 · gepa-ai.github.io
License
The GitHub repository lists the project under the MIT license.Read 9 Oct 2026 · github.com
Agent skill
The repository includes a GEPA Agent Skill for coding agents including Claude Code, Cursor, VS Code/Copilot, Codex, and Gemini CLI.Read 9 Oct 2026 · github.com
Best-fit workloads
GEPA is presented as a fit for expensive rollouts, scarce data, API-only models, or cases where human-readable optimization traces are useful.Read 9 Oct 2026 · gepa-ai.github.io
Budget controls
Users can cap metric calls and configure timeout, no-improvement, score-threshold, signal, file, and composite stop conditions.Read 4 Oct 2026 · gepa-ai.github.io
Built-in adapters
Built-in adapters include Default, Confidence, DSPy Full Program, Generic RAG, MCP, TerminalBench, AnyMaths, and LangChain adapters.Read 4 Oct 2026 · github.com
Comparison caveat
The site says gradient-based methods remain effective when data is abundant and there are 100,000 or more cheap rollouts.Read 9 Oct 2026 · gepa-ai.github.io
Data needs
The FAQ says GEPA can show improvements with as few as three examples and recommends aiming for 30–300 examples for best results.Read 4 Oct 2026 · gepa-ai.github.io
Data scale limit
The site says GEPA can work with as few as three examples and describes 100–500 evaluations as a use case for expensive rollouts.Read 9 Oct 2026 · gepa-ai.github.io
Development
GEPA is developed at UC Berkeley Sky Computing Lab through a research collaboration involving UC Berkeley, MIT, Stanford, and Databricks.Read 9 Oct 2026 · gepa-ai.github.io
Framework flexibility
The optimize_anything API can work with any system through a user-written evaluator, without requiring DSPy or another framework.Read 4 Oct 2026 · gepa-ai.github.io
How it works
It uses LLM reflection on execution traces to diagnose failures, propose targeted changes, and select candidates using a Pareto frontier.Read 9 Oct 2026 · gepa-ai.github.io
Installation
The project’s documented install command is `pip install gepa`.Read 9 Oct 2026 · github.com
Interpretability
GEPA provides human-readable optimization traces that show why prompts changed and can help debug agent behavior.Read 4 Oct 2026 · gepa-ai.github.io
Maker and origin
GEPA is developed at UC Berkeley Sky Computing Lab through a research collaboration between UC Berkeley, MIT, Stanford, and Databricks.Read 4 Oct 2026 · gepa-ai.github.io
Model access
GEPA can optimize API-only models such as GPT, Claude, and Gemini without requiring access to model weights.Read 9 Oct 2026 · gepa-ai.github.io
Notable limitation
GEPA does not inherently optimize for short prompts; its prompts may be longer unless prompt length is included as an optimization objective.Read 4 Oct 2026 · gepa-ai.github.io
Optimization method
It uses execution traces and evaluator-provided diagnostic feedback to guide LLM reflection, targeted mutations, and Pareto-aware candidate selection.Read 4 Oct 2026 · gepa-ai.github.io
Security and compliance
The opened official pages provide no security or compliance certification claims.Read 4 Oct 2026 · gepa-ai.github.io
Supported systems
The project documents adapters for DSPy, RAG, MCP, TerminalBench, and LangChain.Read 9 Oct 2026 · gepa-ai.github.io
Trace feedback
Evaluators can provide error messages, profiling data, and reasoning logs as feedback for optimization.Read 9 Oct 2026 · gepa-ai.github.io
Use cases
The site lists production, coding agent, healthcare, research, and AI safety applications among its GEPA case studies.Read 9 Oct 2026 · gepa-ai.github.io

Plans and prices

GEPA Free Open-source Python package · install with pip · MIT license Read 9 Oct 2026 · github.com

Every GEPA plan and price →

Strong points and limits

Strong points

  • Free, open-source MIT-licensed package
  • Works with API-only GPT, Claude, and Gemini
  • Human-readable traces explain optimization changes
  • Supports multiple adapters and integrations
  • Configurable metric and stopping controls

Limits

  • Does not inherently optimize for short prompts
  • Best-results guidance recommends 30–300 examples
  • Requires a user-written evaluator for arbitrary systems

Who it suits

GEPA suits teams optimizing text-defined systems when rollouts are expensive, data is scarce, models are API-only, or interpretable traces matter. It is also suited to users who can provide an evaluator for their system.

Verdict

GEPA provides broad, evaluator-driven optimization for text-based artifacts and supports self-hosted or API use. Its traces, adapters, and budget controls are useful when interpretability and limited rollout budgets matter, but prompt length requires an explicit objective.

Where Specifiction places it

Alternatives to GEPAAll alternatives →