Privacy choices: we'd like to measure visits anonymously (our own cookie-free counter plus Google Analytics, which sets its own cookies; no ads) and load Beehiiv's newsletter sign-up, which may set its own cookies. Nothing optional runs until you choose. Privacy policy

AgentScout / evals & observability / GEPA
evals & observabilityOpen sourceNewAdded Nov 20, 2026

GEPA logoGEPA

No ratings yet★ Rate this tool

GEPA is an MIT library for optimizing text you can score, including prompts, code and agent architectures, by letting a model reflect on what failed and searching for better variants. The docs site is the project homepage. GitHub listed 7,513 stars on gepa-ai/gepa on 20 Nov 2026.

Updated

✨ Try AgentScout's free AI image generator →

Good for: writing code, building AI agents, research, evaluating agents

Use cases

Get featured

About GEPA

GEPA is listed in our evals & observability category and is aimed at builders. GEPA is an MIT library for optimizing text you can score, including prompts, code and agent architectures, by letting a model reflect on what failed and searching for better variants. The docs site is the project homepage. GitHub listed 7,513 stars on gepa-ai/gepa on 20 Nov 2026. Pricing, as the maker states it: Open source. Maturity: Mature. We last checked the official link on 2026-11-20.

GEPA at a glance

GEPA FAQ

What is GEPA?

GEPA is an MIT library for optimizing text you can score, including prompts, code and agent architectures, by letting a model reflect on what failed and searching for better variants. The docs site is the project homepage. GitHub listed 7,513 stars on gepa-ai/gepa on 20 Nov 2026.

How much does GEPA cost?

GEPA's pricing, as the maker states it: Open source. Open-source tools are free to use, though you may still pay for model API usage or a hosted plan. Check the official site for current plans.

Where can I find GEPA's source code?

On GitHub at gepa-ai/gepa, which had 7,513 stars as of Nov 20, 2026.

What are alternatives to GEPA?

Similar tools include Opik, Codeburn, ReLE Chinese LLM Benchmark, Waku Agent, Superlog and CoreWeave Forge.

Facts from our link-checked listing, the GitHub repository and the official site. See something wrong? Suggest an edit.

GEPA ratings and reviews

No ratings yet. Be the first to review GEPA.

Be the first to rate GEPA: tap 1 to 5 stars below. A written review is optional.

Your rating of GEPA

GEPA alternatives

Other evals & observability tools in our directory. See all evals & observability tools →

Opik logo

Opik

Opik is Comet's Apache-2.0 platform for LLM observability and evaluation: it traces LLM calls, RAG pipelines and agent runs, scores them with built-in and custom metrics (including LLM-as-judge), runs experiments on datasets, manages prompts and monitors production traffic with dashboards. SDKs and integrations cover Python, TypeScript, OpenAI, LangChain, LlamaIndex and other frameworks. The full platform is free to self-host, or you can use Comet's hosted cloud.

Codeburn logo

Codeburn

Free, local tool to track AI coding token usage and cost across 37 tools and agents (Claude Code, Cursor, Codex, Gemini and more), by model, project, and task. npx codeburn

ReLE Chinese LLM Benchmark logo

ReLE Chinese LLM Benchmark

ReLE (formerly CLiB) is NoneLinear's continuously updated benchmark of how well large language models handle Chinese, covering more than 400 commercial and open models such as GPT, Gemini, Claude, Grok, Qwen, DeepSeek, Kimi, GLM and ERNIE. Models are scored across about 300 dimensions in seven domains including education, medicine, finance, law, reasoning and math, instruction following, and agents and tool calling, and the project also publishes a defect library of over two million failure cases.

Waku Agent logo

Waku Agent

Waku Waku! Waku Agent is a local-first AI agent harness you actually own, including loop, memory, eval, all in code built to stay legible as it grows

Superlog logo

Superlog

Superlog runs AI agents on routine engineering work around production. You set up automations with triggers and connectors, for example rating each new Sentry issue, posting a plain-language summary to Slack and opening a pull request with the fix, comparing error rates and latency against their normal range every hour, adding missing logs and spans to hard-to-debug code paths, or answering community questions from your code and docs. You choose the model and agent harness and bring your own tokens or subscriptions. The core is open source (Apache-2.0) with a hosted version.

CoreWeave Forge logo

CoreWeave Forge

One product to run, observe, evaluate and post-train agents (W&B, OpenPipe, sandboxes).