BIRD Mini-Dev - PostgreSQL: leaderboard
Official BIRD Mini-Dev baseline execution accuracy across SQLite, MySQL and PostgreSQL; Overall is their rounded arithmetic mean. Separate from full BIRD Test and New Dev20251106. Source model names retain the baseline run because effort, token budget and dated backend are undisclosed. Empirical upper bound is a reference, not a model. Questions and evaluation code: https://github.com/bird-bench/mini_dev.
Metric: Execution Accuracy (%). Source: bird-bench.github.io. Saturation forecast: Rough model projection: around 2026. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6.1 Sol (Benchmark Sweep Run: source=BIRD MiniDev; config=official baseline; effort=undisclosed) | 89 | |
| 2 | Claude Opus 5.5 (Benchmark Sweep Run: source=BIRD MiniDev; config=official baseline; effort=undisclosed) | 87 | |
| 3 | Claude Sonnet 5.5 (Benchmark Sweep Run: source=BIRD MiniDev; config=official baseline; effort=undisclosed) | 85.6 | |
| 4 | Qwen3.8-Flash-Next (Benchmark Sweep Run: source=BIRD MiniDev; config=official baseline; effort=undisclosed) | 67.8 | |
| 5 | DeepSeek-V4.1-Flash (Benchmark Sweep Run: source=BIRD MiniDev; config=official baseline; effort=undisclosed) | 62.2 | |
| 6 | Claude Haiku 5.5 (Benchmark Sweep Run: source=BIRD MiniDev; config=official baseline; effort=undisclosed) | 58.6 | |
| 7 | GPT-6 Luna (Benchmark Sweep Run: source=BIRD MiniDev; config=official baseline; effort=undisclosed) | 51.8 | |
| 8 | GLM-5.3 (Benchmark Sweep Run: source=BIRD MiniDev; config=official baseline; effort=undisclosed) | 48.4 | |
| 9 | Qwen3.6-35B-A3B (Benchmark Sweep Run: source=BIRD MiniDev; config=official baseline; effort=undisclosed) | 20 |
Interactive version: theaggregate.ai/benchmark?slug=bird-mini-dev-postgresql · How It Works · Data refreshed daily, snapshot 2026-11-06.