BIRD-Interact (a-Interact): leaderboard
Agentic text-to-SQL evaluation where models autonomously interact with databases, planning multi-step SQL operations to fulfill user requests.
Metric: Normalized Reward. Source: github.com. Saturation forecast: Around December 2026. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 25.52 | #98 |
| 2 | Claude Sonnet 4 | 23.28 | #199 |
| 3 | Claude 3.7 Sonnet | 17.45 | #249 |
| 4 | Gemini 2.5 Pro | 17.33 | #152 |
| 5 | O3 Mini | 16.43 | #272 |
| 6 | DeepSeek V3.1 | 13.47 | #264 |
| 7 | Qwen 3 Coder 480B A35B | 10.58 | #279 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=bird-interact-a-interact · How It Works · Data refreshed daily, snapshot 2026-10-20.