Boundary Interaction on Boundary-Aware NL2SQL Synthesized Corpus
0Ambiguity RateGemini3-Pro
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Gemini3-ProModel Category=General-Purpose Baselines2026.01 | 0 | 0 | 10.34 | 0 | 29.9 | |
| DeepSeek-V3.2Model Category=General-Purpose Baselines2026.01 | 0 | 1.69 | 17.19 | 0 | 29.17 | |
| Qwen3-1.7B (Base)Model Category=Ours, Training Stage=Base2026.01 | 0 | 0 | 0 | 0 | 8.47 | |
| GPT-4oModel Category=General-Purpose Baselines2026.01 | 2.15 | 0 | 34.38 | 2.9 | 18.59 | |
| GPT-5Model Category=General-Purpose Baselines2026.01 | 5.56 | 13.79 | 70.49 | 48.24 | 46.01 | |
| Claude-4.5-SonnetModel Category=General-Purpose Baselines2026.01 | 16.13 | 18.64 | 87.5 | 17.75 | 48.02 | |
| Qwen3-1.7B (SFT)Model Category=Ours, Training Stage=SFT2026.01 | 68.75 | 42.37 | 86.12 | 55.8 | 67.55 | |
| BAR-SQLModel Category=Ours, Training Stage=GRPO2026.01 | 94.51 | 93.31 | 92.27 | 92.36 | 91.48 |