FinMTM - Self-Correction: leaderboard
Metric: FinMTM open-ended self-correction (L3: adversarial robustness and logical consistency) score (0-100) over 1,210 multi-turn financial sessions (about 4.25 turns each): half the mean LLM-judge rating of each turn on visual precision, financial logic, data accuracy, cross-modal verification and temporal awareness, half a session-level checklist for the task type; higher is better. Source: arxiv.org. Saturation forecast: Around December 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 58.8 | #77 |
| 2 | GPT-5 | 56.9 | #91 |
| 3 | Gemini 3 Flash | 55.4 | #93 |
| 4 | Qwen 3 VL 235B A22B Instruct | 54.5 | #264 |
| 5 | O3 | 52.8 | #121 |
| 6 | Qwen 3 VL 235B A22B (Thinking) | 52.5 | #228 (Qwen 3 VL 235B A22B) |
| 7 | GLM-4.5V | 51.1 | #339 |
| 8 | Qwen 3 VL 32B Instruct | 50.8 | #276 |
| 9 | GPT-4o | 46.2 | #333 |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 44.2 | #338 (Qwen 3 VL 30B A3B) |
| 11 | InternVL3-78B | 43.6 | #345 |
| 12 | Qwen 3 VL 32B (Thinking) | 43.5 | #287 (Qwen 3 VL 32B) |
| 13 | Qwen 2.5 VL 7B Instruct | 43.1 | #643 |
| 14 | Qwen 3 VL 30B A3B Instruct | 42.5 | #365 |
| 15 | Qwen 3 VL 4B (Thinking) | 42.5 | #471 (Qwen 3 VL 4B) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=finmtm-self-correction · How It Works · Data refreshed daily, snapshot 2026-10-11.