FinToolBench - Soft Score: leaderboard
Metric: Soft Score (0-1, times 100): mean GPT-5.1 judgment of the final answer over all questions (numeric and choice answers scored 1 or 0, structured-analysis and other answers 1, 0.5 or 0; three judging repeats averaged), on the 295 tool-required financial questions (166 single-tool, 129 multi-tool) over 760 free-tier RapidAPI and AkShare tools, with the model as the planner of the authors' FATR agent (BGE-M3 retrieval of candidate tools, finance-attribute tool cards, ReAct loop of at most 5 tool steps); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 82.2 | #54 |
| 2 | Claude Sonnet 4.5 | 71.19 | #138 |
| 3 | Grok 3 | 68.47 | #296 |
| 4 | GPT-5.4 | 62.54 | #76 |
| 5 | Seed-1.6 | 46.27 | #257 |
| 6 | Qwen 3 8B | 40.4 | #667 |
| 7 | GLM-4.7 Flash | 37.91 | #496 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=fintoolbench-soft-score · How It Works · Data refreshed daily, snapshot 2026-10-11.