UniQL - Spark: leaderboard
Metric: Execution accuracy (%) on the Spark dialect of UniQL (1,534 BIRD development-set questions with human-verified SQL in 16 aligned dialects): given the target dialect, schema and question, the model writes one query whose execution result must match the gold result; inference-only, one run; higher is better. Source: arxiv.org. Saturation forecast: Around September 2028. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro | 56.65 |
| 2 | Claude Sonnet 4.5 | 56.13 |
| 3 | GPT-5.1 Codex | 53.65 |
| 4 | GPT-5 Mini | 52.93 |
| 5 | DeepSeek V4 Flash | 51.56 |
| 6 | Qwen 3 32B | 49.22 |
| 7 | Qwen 3 8B | 47.46 |
| 8 | Qwen 3 4B | 43.94 |
| 9 | Llama 3 70B Instruct | 42.89 |
| 10 | GPT-3.5 Turbo | 39.37 |
| 11 | Qwen 3 1.7B | 35.27 |
| 12 | Llama 3 8B Instruct | 23.34 |
Interactive version: theaggregate.ai/benchmark?slug=uniql-spark · How It Works · Data refreshed daily, snapshot 2026-09-29.