Text-to-SQL under RBAC - LiveSQLBench: leaderboard

Access-controlled F1 on the LiveSQLBench split of the role-based-access text-to-SQL evaluation. Measures answering database questions while respecting the user's access permissions. Spider and BIRD are separate boards.

Metric: Access-Controlled F1 (%, LiveSQLBench). Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 4 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-561.15#180
2GPT-5 Mini58.49#365
3Claude Sonnet 4.555.75#262
4DeepSeek V3.2Thinking48.8#423
5Gemini 2.5 Flash44.89#469
6DeepSeek V3.2 Coder38.47
7Qwen 2.5 14B Instruct37.46#1785
8Gemma 3 27B34.12#1669
9Qwen 2.5 Coder 7B Instruct31.86#3987
10GPT-4o Mini30.46#1678
11Snowflake-R1-7B29.33
12Gemma 3 4B28.89#5301

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=text-to-sql-under-rbac-livesqlbench · How It Works · Data refreshed daily, snapshot 2026-11-20.