Skip to content

Data Extraction Race

Which model reads a document best?

Pick a document and a few models. Every model extracts the same fields; we score each answer against the ground truth — no LLM judge, just exact and fuzzy field matching.

Share this challenge

What this is

Public game

Anyone can play, free. Pick a real document, run the same extraction across the models you choose, and get an objective score — exact and fuzzy field matching against the ground truth, no LLM judge.

Admin arena

A separate staff tool for authoring and curating custom matches. You don’t need it to play here — it’s how new public games get built.

Pick a document

Pick models (2–6) · 3/6

We pre-select a few cheap models plus one frontier model so you can see the price/quality gap.

gpt-oss-20bOVH AI Endpoints (GRA) · $0.11/$0.39 per 1M100%Mistral-7B-Instruct-v0.3OVH AI Endpoints (GRA) · $0.26/$0.26 per 1M99%Claude Fable 5Anthropic · $13.00/$65.00 per 1M96%

Verifying you are human…

Estimated: up to €0.07 First 5 games/day are free
5 free games left todayHow scoring works