Datasets:

Hawky-ai
/

pm-agi-benchmark

license: mit
task_categories:
  - question-answering
  - text-generation
language:
  - en
tags:
  - performance-marketing
  - meta-ads
  - google-ads
  - benchmark
  - evaluation
  - llm-evaluation
  - advertising
pretty_name: PM-AGI Benchmark
size_categories:
  - n<1K

PM-AGI Benchmark 🎯

The first open-source LLM benchmark for Performance Marketing.

Developed by hawky.ai — evaluating how well LLMs reason, plan, and act in real-world Meta Ads and Google Ads scenarios.

Dataset Summary

PM-AGI contains 100 expert-crafted questions across 4 categories of performance marketing knowledge:

Category	Questions	Focus
Meta Ads	30	Campaign structure, targeting, bidding, creative, CAPI, measurement
Google Ads	30	Search, Smart Bidding, PMax, Quality Score, attribution
Critical Thinking	20	Data interpretation, budget decisions, competitive analysis
Action-Based	20	Scenario troubleshooting, optimization, scaling

Question Types

MCQ (63 questions) — Single correct answer, scored 1.0 or 0.0
Action-Based (37 questions) — Open scenario evaluated by LLM judge (0.0–1.0)

Difficulty Distribution

Easy: 9 questions
Medium: 50 questions
Hard: 41 questions

Usage

from datasets import load_dataset

ds = load_dataset("Hawky-ai/pm-agi-benchmark")
print(ds["test"][0])

Evaluate a Model

git clone https://github.com/Hawky-ai/pm-AGI
cd pm-agi-benchmark
pip install -r requirements.txt
python evaluate.py --model gpt-4o --provider openai --api-key YOUR_KEY

Leaderboard

🏆 Live Leaderboard

Citation

@misc{pmagi2025,
  title={PM-AGI: A Performance Marketing Benchmark for Large Language Models},
  author={hawky.ai},
  year={2025},
  url={https://huggingface.co/datasets/Hawky-ai/pm-agi-benchmark}
}

License

MIT — see LICENSE