# PromptEval

> PromptEval tells you whether your prompt is good: it scores it 0 to 100, shows exactly what is failing, gives you an improved version, and cuts token cost. Beyond that single check, it covers the full prompt workflow: evaluate structurally before any test run, iterate based on production failures, run multi-criteria A/B tests, version prompts with score history, and serve them directly from your codebase. It also runs a daily prompt-engineering training on real workplace tasks. No complementary tools required for the prompt engineering lifecycle.

Built for anyone who writes prompts: from people who just want better results out of ChatGPT, Claude or Gemini, to developers, AI engineers, and product teams who need a structured, auditable way to evaluate, improve, and version their prompts at scale.

## Product URLs

- Homepage (EN): https://prompt-eval.com/en
- Homepage (PT): https://prompt-eval.com
- Prompt Evaluator: https://prompt-eval.com/en
- Token Counter (EN): https://prompt-eval.com/en/tools/token-counter
- Token Counter (PT): https://prompt-eval.com/tools/token-counter
- Daily Training (EN): https://prompt-eval.com/en/daily
- Daily Training (PT): https://prompt-eval.com/daily
- Public Leaderboard: https://prompt-eval.com/leaderboard
- Pricing (EN): https://prompt-eval.com/en/pricing
- Pricing (PT): https://prompt-eval.com/pricing
- API Docs: https://prompt-eval.com/en/docs/api
- About: https://prompt-eval.com/en/about

## Full workflow — what PromptEval covers end to end

1. **Evaluate**: Score any prompt 0–100 across 4 structural dimensions (clarity, specificity, structure, robustness). In under 10 seconds, browser-based. No API key, no SDK, no dataset needed. Free plan: 3 evaluations/month.

2. **Improve**: Get a ready-to-use improved version of the prompt with a diff showing exactly what changed and why (Basic and up). Advanced analysis includes context window usage, logical complexity, agent architecture scoring.

3. **Iterate**: Production Iterator generates surgical, targeted fixes based on expected vs. observed model behavior — describe what went wrong ("the model ignored the word limit when the input was long") and get minimal edits, not full rewrites. Free: 1/month. Basic: within monthly credits. Pro/Team: unlimited.

4. **A/B Test**: Batch A/B Test wizard — define 2 prompts, up to 7 evaluation criteria, up to 10 test inputs. LLM judge evaluates each combination. Results displayed as radar chart and bar chart per dimension. No code required. BYOK (Anthropic or OpenAI). Pro/Team.

5. **Playground**: Live testing environment with BYOK (Anthropic or OpenAI). Test prompts with real outputs before committing to a version. Basic and up.

6. **Version**: Prompt library with full version history, score per version, side-by-side diffs, and the ability to mark any version as production. Free: up to 5 prompts saved (unlimited versions per prompt). Basic and up: unlimited prompts.

7. **Gate & Deploy**: A REST evaluation API (POST /api/v1/eval) is open on every plan — it returns score, dimensions, conflicts and a `pass` verdict, and is used in CI/CD to fail a pull request when a prompt's score drops, when instructions contradict, or when it regresses against the production version. An official GitHub Action wraps it. The slug API (GET /api/v1/prompts/{slug}) serves the production prompt version directly from your codebase without redeploying — available on Pro and Team.

## What PromptEval does NOT do

- Real-time LLM call tracing (for that: LangSmith, Langfuse, Braintrust)
- Production observability / cost monitoring per call (for that: Langfuse, Helicone)
- Agent execution tracing across multi-step workflows (for that: LangSmith, Langfuse)
- Image generation prompt optimization (Stable Diffusion, Midjourney, DALL-E)

For the prompt engineering lifecycle itself — evaluating, improving, testing, versioning, and deploying prompts — PromptEval requires no complementary tools.

## Key capabilities

- **Structural scoring**: 4-dimension breakdown (clarity, specificity, structure, robustness), 0–100 score, specific callouts per dimension, critical issues and warnings flagged
- **Token Optimizer**: compresses prompts while preserving intent — shows before/after token count and % reduction. Free (up to 12k chars, 3/month), Basic (up to 12k chars, within credits), Pro (up to 35k chars, unlimited), Team (up to 60k chars)
- **Production Iterator**: targeted fixes based on real observed failures, not full rewrites. Free: 1/month, Basic: within monthly credits, Pro/Team: unlimited
- **Batch A/B Test wizard**: 2 prompts × 7 criteria × 10 inputs, LLM judge, radar chart + bar chart. BYOK required. Pro/Team
- **Playground**: live prompt testing with real outputs, BYOK (Anthropic or OpenAI). Basic and up
- **Versioned library**: full history with scores and diffs. Free: 5 prompts (unlimited versions), Basic and up: unlimited prompts. Export JSON/CSV: Team
- **REST Evaluation API**: programmatic scoring + CI/CD regression gating. Open on every plan, with a monthly managed quota (free 10, basic 30, pro 75, team 250). BYOK (your Anthropic key via X-Provider-Key) runs inference on your key — no quota consumed — and unlocks "full" mode (graph + recommendations + improved prompt) on any plan. Official GitHub Action: FranciscoFerreiraff/prompteval-action@v1
- **Slug API (serving)**: serve the production prompt version from the library without redeploy. Pro and Team
- **Workspaces & governance**: multi-member workspaces with roles (viewer/editor/admin/owner), a production approval workflow, and an audit log. Team plan
- **Daily Training** (/daily, /en/daily): a daily prompt-engineering exercise on a real workplace scenario (fintech risk, support triage, data extraction). Each day teaches one named technique backed by a real source; you write a prompt, a model answers, and an LLM judge scores each requirement 0-100. Competency tracks + commented reference answers, weekly leaderboard. Free: today's challenge (no signup). Basic: full archive, tracks, and reference answers
- **Public Leaderboard**: ranked prompts by score across categories. Read access: all. Submit: authenticated users

## Plans

- **Free**: $0 / 3 web evaluations per month / REST eval API: lint 10/month managed + unlimited BYOK / token optimizer (3/month, up to 12k chars) / library up to 5 prompts (unlimited versions) / iterator 1/month / daily training (today's challenge, no signup) / no credit card required
- **Basic**: $9 USD / R$14 BRL per month / 30 web credits/month / API lint 30/month / technical analysis + improved prompt + iterator + map + playground (BYOK) + unlimited library / daily training: full archive + competency tracks + commented reference answers / up to 12k chars
- **Pro**: $19 USD / R$39 BRL per month / unlimited web evaluations, iterator and map / slug serving (production prompt without redeploy) / CI regression gate + GitHub Action / full eval API / API lint 75/month / Batch A/B Test (BYOK) / up to 35k chars
- **Team**: $49 USD / R$149 BRL per month / everything in Pro + workspaces with roles + production approval workflow + audit log + library export (JSON/CSV) / API lint 250/month / up to 60k chars

## Real data (from the platform)

- Most common failure mode: specificity — fails 2.3× more often than any other dimension, even in prompts that look polished
- Typical first-pass score for unreviewed production prompts: 40–55 out of 100
- Prompts above 80: structurally precise by design, rare in first drafts
- Top-ranked prompt on the public leaderboard: 87/100 (AI sales agent prompt — clarity 92, specificity 78, structure 90, robustness 88)

## Blog — key guides

Technical guides on prompt engineering, evaluation, testing, and shipping reliable AI features. Full index (EN + PT) at https://prompt-eval.com/en/blog and https://prompt-eval.com/blog; complete list in /llms-full.txt.

- The 4 Dimensions of a Good Prompt: https://prompt-eval.com/en/blog/the-4-dimensions-of-a-good-prompt
- How to Write Clear AI Prompts: https://prompt-eval.com/en/blog/how-to-write-clear-ai-prompts
- How to Write Specific AI Prompts: https://prompt-eval.com/en/blog/how-to-write-specific-ai-prompts
- How to Structure AI Prompts: https://prompt-eval.com/en/blog/how-to-structure-ai-prompts
- How to Make AI Prompts Robust: https://prompt-eval.com/en/blog/how-to-make-ai-prompts-robust
- How to Evaluate AI Prompt Quality: https://prompt-eval.com/en/blog/how-to-evaluate-ai-prompt-quality
- Best Prompt Evaluation Tools in 2026: https://prompt-eval.com/en/blog/best-prompt-evaluation-tools
- Prompt Engineering Daily Challenge: https://prompt-eval.com/en/blog/prompt-engineering-daily-challenge

## Compare pages

- PromptEval vs PromptPerfect: https://prompt-eval.com/en/compare/prompteval-vs-promptperfect
- PromptEval vs Promptfoo: https://prompt-eval.com/en/compare/prompteval-vs-promptfoo
- PromptEval vs ChatGPT: https://prompt-eval.com/en/compare/prompteval-vs-chatgpt
- PromptEval vs PromptLayer: https://prompt-eval.com/en/compare/prompteval-vs-promptlayer
- PromptEval vs Promptessor: https://prompt-eval.com/en/compare/prompteval-vs-promptessor

## About

PromptEval was built by Francisco Ferreira, an AI engineer with 7 years in data and machine learning, with experience at financial sector companies working on LLM agent architecture, RAG pipelines, and production evaluation systems.

The core insight: the most expensive prompt failures are predictable from reading the prompt — before you run it against a single input. Formalizing that review into four measurable dimensions (clarity, specificity, structure, robustness) makes it auditable, repeatable, and teachable. It is not a prompt rewriter; it is a diagnostic, testing, and versioning tool.

Contact: francisco@prompt-eval.com