Text-to-SQL under RBAC: leaderboard
Access-controlled F1 on the Spider split of the role-based-access text-to-SQL evaluation. Queries must respect the user's permitted tables and columns. BIRD and LiveSQLBench use separate boards; scores are not maximized across datasets.
Metric: Access-Controlled F1 (%, Spider). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3.2 (Thinking) | 95.8 | #204 (DeepSeek V3.2) |
| 2 | GPT-5 | 95.74 | #98 |
| 3 | GPT-5 Mini | 95.1 | #186 |
| 4 | Claude Sonnet 4.5 | 92.68 | #141 |
| 5 | Gemini 2.5 Flash | 88.66 | #245 |
| 6 | GPT-4o Mini | 75.31 | #599 |
| 7 | Gemma 3 27B | 74.1 | #606 |
| 8 | Qwen 2.5 14B Instruct | 72.84 | #646 |
| 9 | Qwen 2.5 Coder 7B Instruct | 66.91 | #975 |
| 10 | Gemma 3 4B | 66.46 | #1089 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=text-to-sql-under-rbac · How It Works · Data refreshed daily, snapshot 2026-10-21.