# FinanceBench leaderboard

Open-book questions about US public-company filings (10-Ks, 10-Qs, 8-Ks, earnings reports). The model must find and compute answers from the filing, usually through retrieval or long context.

## FinanceBench: % correct, no filing given (150-question sample) (higher is better, 5 models)

| Rank | Model | Provider | Score | Metric | Setting | Reported by | Source | URL | Date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | Numa 1 | Moneda | 67 | % correct, no filing given (150-question sample) | Finds its own filings, no documents given; graded by a two-judge panel (partly right counts half) | Run by Moneda | Moneda run, 2026-11-02 | https://benchmarks.finance/method/ | 2 Nov 2026 |
| 2 | GPT-5.4 | OpenAI | 63.7 | % correct, no filing given (150-question sample) | Built-in web search, no documents given; same 150 questions and two-judge panel as Numa 1 (partly right counts half) | Run by Moneda | Moneda run, 2026-11-02 | https://benchmarks.finance/method/ | 2 Nov 2026 |
| 3 | Claude Opus 5.5 | Anthropic | 60.3 | % correct, no filing given (150-question sample) | Built-in web search, no documents given; same 150 questions and two-judge panel as Numa 1 (partly right counts half) | Run by Moneda | Moneda run, 2026-11-02 | https://benchmarks.finance/method/ | 2 Nov 2026 |
| 4 | GPT-4-Turbo | OpenAI | 19 | % correct, no filing given (150-question sample) | Shared vector store (all filings) | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2311.11944v1 | 20 Nov 2023 (checked 2 Nov 2026) |
| 4 | Llama2 | Meta | 19 | % correct, no filing given (150-question sample) | Shared vector store (all filings) | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2311.11944v1 | 20 Nov 2023 (checked 2 Nov 2026) |

## FinanceBench: % correct, the right filing given (150-question sample) (higher is better, 3 models)

| Rank | Model | Provider | Score | Metric | Setting | Reported by | Source | URL | Date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | GPT-4-Turbo | OpenAI | 85 | % correct, the right filing given (150-question sample) | Oracle (evidence pages given) | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2311.11944v1 | 20 Nov 2023 (checked 2 Nov 2026) |
| 2 | Claude2 | Anthropic | 76 | % correct, the right filing given (150-question sample) | Long context (filing in context) | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2311.11944v1 | 20 Nov 2023 (checked 2 Nov 2026) |
| 3 | Llama2 | Meta | 41 | % correct, the right filing given (150-question sample) | Single vector store (one per filing) | Reported by the authors | Paper Table 2 | https://arxiv.org/html/2311.11944v1 | 20 Nov 2023 (checked 2 Nov 2026) |

Scores are never combined across benchmarks. Each row names its metric, source and date.

## About this benchmark

- Maker: Patronus AI, with Contextual AI and Stanford University
- Year: 2023
- Use case: Financial filings Q&A
- Size: 10,231 questions over 361 public filings
- Public split: 150 annotated questions, with source PDFs and the paper's human-graded model outputs. The full set is available on request from Patronus AI.
- Grading: Humans label each answer correct, incorrect or failed to answer. Small unit and rounding slips are allowed.
- Licence: CC-BY-NC-4.0 (data, per Hugging Face card)
- Submission: via-maker
- Products allowed: Not stated
- Latest version: arXiv v1 (2023-11-20); HF dataset last modified 2024-11-17
- What it does not cover: No live data, no multi-step agent workflows, no non-US filings, and no grading of the reasoning path beyond the final answer.
- Last checked: 1 Nov 2026
- Links: paper https://arxiv.org/abs/2311.11944, data https://huggingface.co/datasets/PatronusAI/financebench, code https://github.com/patronus-ai/financebench
- Directory page: https://benchmarks.finance/benchmarks/financebench/
- Leaderboard: https://benchmarks.finance/leaderboards/financebench/

Source: https://benchmarks.finance/leaderboards/financebench/

Created and maintained by Moneda.