# PRBench Finance

High-stakes professional reasoning in finance conversations, graded against expert-written rubrics. Finance is one half of PRBench; the other half is Legal.

## Facts

- Maker: Scale AI
- Year: 2025
- Use case: Financial advice and safety
- Size: 600 Finance tasks (plus a 300-task Finance-Hard subset). Across Finance and Legal: 1,100 tasks.
- Public split: All tasks and rubrics on Hugging Face (not gated)
- Grading: Weighted rubric score clipped to 0-1, with 10-30 expert criteria per task. The board shows it on a 0-100 scale.
- Licence: CC-BY-4.0 (data)
- Submission: Not stated
- Products allowed: Not stated
- Latest version: HF last modified 2026-06-24
- What it does not cover: No tool use, live data retrieval or numeric spreadsheet work.
- Last checked: 1 Nov 2026
- Links: paper https://arxiv.org/abs/2511.11562, data https://huggingface.co/datasets/ScaleAI/PRBench, leaderboard https://labs.scale.com/leaderboard/prbench-finance
- Directory page: https://benchmarks.finance/benchmarks/scale-prbench-finance/
- Leaderboard: https://benchmarks.finance/leaderboards/scale-prbench-finance/

## Results

### PRBench Finance: weighted rubric score (0-100) (higher is better, 10 models)

| Rank | Model | Maker | Score | Metric | Setting | Reported by | Source | URL | Date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | Muse Spark 1.3 | Meta | 59.5 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |
| 2 | Muse Wonder 1.0 | Meta | 58.7 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |
| 3 | Muse Spark 1.1 | Meta | 55.0 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |
| 4 | Claude Fable 5 | Anthropic | 53.9 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |
| 5 | claude-opus-4-6 (Non-Thinking) | Anthropic | 53.3 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |
| 6 | Muse Spark | Meta | 52.4 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |
| 7 | gpt-5 | OpenAI | 51.3 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |
| 8 | gpt-5-pro | OpenAI | 51.1 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |
| 9 | Fable 5.1 | Anthropic | 50.8 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |
| 10 | gpt-5.6-sol (max) | OpenAI | 50.5 | weighted rubric score (0-100) |  | Official board | Scale Labs leaderboard | https://labs.scale.com/leaderboard/prbench-finance | date not stated |


Source: https://benchmarks.finance/benchmarks/scale-prbench-finance/

Created and maintained by Moneda.