FinInteract: leaderboard
FinInteract (Wang et al., 2026) tests whether LLM agents resolve deliberately ambiguous financial questions over SEC and Chinese A-share filings by asking a simulated user the right yes-or-no clarifying question before answering. Accuracy in percent against the intended interpretation over all 173 English and Chinese instances, graded by GPT-4o-mini with a 1% numeric tolerance; higher is better.
Metric: Accuracy (%; all 173 bilingual instances; ReAct agent in the standard full-interaction mode (A+S+I): it may search the filing corpus, ask the simulated user (GPT-5, answering yes, no or I don't know from the hidden disambiguating context) yes-or-no questions, and answer; a final answer counts as correct only if it matches the intended interpretation, graded by GPT-4o-mini with a 1% numeric tolerance). Source: arxiv.org. Saturation forecast: Not forecast. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 35B A3B | 28.9 | #480 |
| 2 | Qwen 3 8B | 24.9 | #1838 |
| 3 | Qwen 3 14B | 22.5 | #1235 |
| 4 | GPT-5 | 20.2 | #163 |
| 5 | Qwen 3 4B | 12.1 | #2297 |
| 6 | Qwen 3 30B A3B | 11.6 | #1124 |
| 7 | Qwen 3 32B | 8.4 | #960 |
| 8 | GPT-4o | 4.6 | #741 |
| 9 | GPT-5 Mini | 0 | #349 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=fininteract · How It Works · Data refreshed daily, snapshot 2026-11-12.