Get in touch

Collaborate

Questions about Finance Benchmark, model evaluations, or collaboration — reach out via LinkedIn or the form below.

Connect on LinkedIn for benchmark updates, finance AI discussion, or direct messages.

Give feedback

Evaluate your model

Finance Benchmark is open and reproducible. Run the harness locally, then publish results to appear on the leaderboard.

1. Install the harness

pip install -e .
cp .env.example .env
# Add API keys for your provider

2. Run evaluation

finbench run --model openai/gpt-5.4 --tasks all --runs 3
# Or evaluate the full active set:
finbench run-all --publish

3. Publish results

finbench publish results/<model>_<timestamp>.json

Requires SUPABASE_URL and SUPABASE_SERVICE_ROLE_KEY in .env.

4. Share your run

Result JSON files include per-attempt responses for full transparency. Upload or link your result file in a pull request if you want it included in the public results/ directory for drill-down on the site.

Supported providers

OpenAI, Anthropic, Google Gemini, xAI, and DeepSeek. Use finbench list-models to see the catalog.

See Methodology for scoring rules and held-out task policy.