Vals AI CorpFin v2: leaderboard
Questions about long corporate credit agreements, covering extraction of terms and numbers, interpretation, numeric reasoning, cross-references between sections and finance jargon. The held-out test set holds 858 questions over 43 agreements, and pages of the agreement are supplied in context in three window variants. Claude 4.5 Sonnet grades the answers at temperature zero, and higher is better.
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Around January 2028. 134 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 73.19 | #12 |
| 2 | Claude Fable 5 (Max) | 71.83 | #16 (Claude Fable 5) |
| 3 | Kimi K3 | 71.56 | #30 |
| 4 | Muse Spark 1.1 (xHigh) | 71.29 | #52 (Muse Spark 1.1) |
| 5 | Muse Spark 1.2 (xHigh) | 70.94 | #62 (Muse Spark 1.2) |
| 6 | Inkling Small | 69.62 | #138 |
| 7 | Inkling | 68.57 | #126 |
| 8 | Grok 4.3 | 68.53 | #188 |
| 9 | GPT-5.5 (xHigh) | 68.42 | #31 (GPT-5.5) |
| 10 | Kimi K2.5 (Thinking) | 68.26 | #144 (Kimi K2.5) |
| 11 | MiniMax-M3 | 68.1 | #136 |
| 12 | Qwen 3 Max (2026-01-23) | 68.03 | #155 |
| 13 | Grok 4.5 (High) | 67.41 | #48 (Grok 4.5) |
| 14 | Claude Opus 4.6 (Adaptive Reasoning, Max Effort) | 67.02 | #63 (Claude Opus 4.6) |
| 15 | Claude Sonnet 5 (Max) | 66.98 | #57 (Claude Sonnet 5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-corpfin-v2 · How It Works · Data refreshed daily, snapshot 2026-10-20.