V-FiLLM - Simplified Statements: leaderboard
V-FiLLM (arXiv 2608.11047v1, Table 2): verified financial reasoning, where each question is generated from an executable computation tree over statement values so its answer is deterministic, on 250 mixed-depth questions. Accuracy with simplified spreadsheet statements, which admit deeper reasoning chains, single-turn, percent. Higher is better.
Metric: Accuracy (%; single-turn). Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 4 31B (IT) | 97.6 | #178 |
| 2 | DeepSeek V4 Flash | 95.6 | #155 |
| 3 | Qwen 3.7 Plus | 94.4 | #98 |
| 4 | GPT-OSS-120B | 93.6 | #346 |
| 5 | Llama 3.3 70B Instruct | 80.8 | #541 |
| 6 | Qwen 3.5 9B | 55.2 | #391 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=v-fillm-simplified-statements · How It Works · Data refreshed daily, snapshot 2026-11-05.