# CorpFin v2 leaderboard

Question answering over long credit agreements (often more than 200 pages). The same questions are asked with exact pages, about 80 pages of shared context, or the most context that fits.

## CorpFin v2: accuracy % (higher is better, 10 models)

| Rank | Model | Maker | Score | Metric | Setting | Reported by | Source | URL | Date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | Claude Opus 5 | Anthropic | 73.2 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |
| 2 | Claude Fable 5 | Anthropic | 71.8 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |
| 3 | Kimi K3 | Moonshot AI | 71.6 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |
| 4 | Muse Spark 1.1 | Meta | 71.3 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |
| 5 | Muse Spark 1.2 | Meta | 70.9 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |
| 6 | Inkling Small | Not stated | 69.6 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |
| 7 | Inkling | Not stated | 68.6 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |
| 8 | Grok 4.3 | xAI | 68.5 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |
| 9 | GPT 5.5 | OpenAI | 68.4 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |
| 10 | Kimi K2.5 | Moonshot AI | 68.3 | accuracy % |  | Official board | Vals.ai leaderboard (archived) | https://www.vals.ai/benchmarks/corp_fin_v2 | 12 Aug 2026 |

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

## About this benchmark

- Maker: Vals AI
- Year: Not stated
- Use case: Financial filings Q&A
- Size: 1,218 questions from 61 documents (20 public validation, 340 private validation, 858 test)
- Public split: 20 questions from 1 document, available on request
- Grading: LLM-as-judge (Claude Sonnet 4.5 at temperature 0, which replaced Sonnet 3.5 on 2025-11-17)
- Licence: proprietary
- Submission: none
- Products allowed: Not stated
- Latest version: v2 (archived; last updated 2026-08-12)
- What it does not cover: Archived as saturated. Reading comprehension only: no tools, calculation workflows or market data.
- Last checked: 1 Nov 2026
- Links: site https://www.vals.ai/benchmarks/corp_fin_v2, leaderboard https://www.vals.ai/benchmarks/corp_fin_v2
- Directory page: https://benchmarks.finance/benchmarks/vals-corpfin-v2/
- Leaderboard: https://benchmarks.finance/leaderboards/vals-corpfin-v2/

Source: https://benchmarks.finance/leaderboards/vals-corpfin-v2/

Created and maintained by Moneda.