Benchmax
Benchmarks, built and fixed in the open.
Every case visible. Every label disputable with the page it came from. Every score regraded when the gold changes, never rerun. Fork any benchmark and keep one of your own with the same tool.
Public benchmarks
PDF extraction
Schema-guided extraction on an attributed ExtractBench subset: ten documents, gold adjudicated in the open.
Top system · value f1
0.989
Codex · GPT-5.6 Sol · medium effort
Run it on your documents
Fork a benchmark or start from your own schema. Add PDFs, run any system, confirm gold from the page, fix labels with a reason. Files in your repository, nothing uploaded.
pip install benchmax
benchmax init my-invoices --schema invoice.json
cd my-invoices && benchmax add ./my-pdfs
benchmax run --target extract.py:extract
benchmax serveFix a benchmark
Think a label is wrong? Open the document, click the field, file the dispute with the page evidence. Accepted disputes become a new gold version and every saved output is regraded, never rerun.