PremiseSQL - Implicit prompting - Flaw detection: leaderboard
PremiseSQL premise-critique evaluation over Spider and BIRD Text-to-SQL queries with flawed-premise variants and their paired originals (3,600 requests per model and prompting setting), temperature 0 and default API configuration, two-judge labels. Implicit prompting: the model is told that the question may contain errors and asked to assess its validity before answering. DET: share of flawed queries for which the model detects a premise flaw. Fraction from 0 to 1. Model labels as printed. Metric: DET (fraction). Higher is better. Source: https://arxiv.org/html/2610.22158v1
Metric: DET (fraction). Source: arxiv.org. Saturation forecast: Not forecast. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3.2 | 0.9 | #202 |
| 2 | Qwen 3.5 35B A3B | 0.78 | #254 |
| 3 | Qwen 3.5 397B A17B | 0.72 | #145 |
| 4 | GPT-6.1 Sol | 0.72 | #7 |
| 5 | Qwen 3.5 122B A10B | 0.71 | #171 |
| 6 | Gemini 3.1 Pro (Preview) | 0.71 | #57 |
| 7 | GPT-5.4 | 0.7 | #79 |
| 8 | Seed 2.0 Pro | 0.68 | #117 |
| 9 | Claude Opus 5.5 | 0.67 | #4 |
| 10 | GPT-6 Luna | 0.65 | #52 |
| 11 | GPT-5.6 Luna | 0.63 | #68 |
| 12 | DeepSeek V4.1 Flash | 0.63 | #35 |
| 13 | Llama 4 Maverick | 0.59 | #465 |
| 14 | Claude Opus 4.6 | 0.58 | #62 |
| 15 | Claude Sonnet 4.6 | 0.48 | #90 |
No result here: #6 GPT-6 Astra, #8 Gemini 4 Argon, #9 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=premisesql-implicit-prompting-flaw-detection · How It Works · Data refreshed daily, snapshot 2026-10-18.