Attribution Consistency and Downstream Performance on FinQA-TAS
76.9F1 ScoreGPT-5.1
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5.1Size=Closed, Prompting Strategy=Prompting2026.04 | 76.9 | — | — | — | |
| DeepSeek-V3.2Size=671B, Prompting Strategy=Prompting2026.04 | 76 | — | — | — | |
| ReTASSize=4B2026.04 | 72.1 | 71.2 | 12.4 | 5.4 | |
| QwQ-32BSize=32B, Prompting Strategy=Reflection: Dual View2026.04 | 71 | 54.9 | 18.1 | 14.7 | |
| QwQ-32BSize=32B, Prompting Strategy=Prompting2026.04 | 68.9 | — | — | — | |
| QwQ-32BSize=32B, Prompting Strategy=Reflection: Single View2026.04 | 68.4 | 53.1 | — | — | |
| Qwen3-30B-A3BSize=30B, Prompting Strategy=Reflection: Dual View2026.04 | 66.5 | 52.9 | 20.1 | 13.5 | |
| GLM-4.6Size=9B, Prompting Strategy=Reflection: Dual View2026.04 | 66.3 | 43.1 | 52.7 | 24.8 | |
| GLM-4.6Size=9B, Prompting Strategy=Reflection: Single View2026.04 | 64.9 | 43.7 | — | — | |
| Qwen3-30B-A3BSize=30B, Prompting Strategy=Reflection: Single View2026.04 | 63.6 | 49.8 | — | — | |
| Qwen3-30B-A3BSize=30B, Prompting Strategy=Prompting2026.04 | 61 | — | — | — | |
| GLM-4.6Size=9B, Prompting Strategy=Prompting2026.04 | 60.4 | — | — | — |