ProcArena - Oracle - Interactive - From-Scratch Synthesis: leaderboard
ProcArena tests PL/SQL synthesis and editing on PostgreSQL and Oracle by executing generated code against held-out calls and checking outputs and persistent database state. Direct generation and interactive tool-assisted development are separate evaluation modes. This series reports ProcArena - Oracle - Interactive - From-Scratch Synthesis from 2609.06527v1, S5.T6: Execution Accuracy (%). Higher is better.
Metric: Execution Accuracy (%). Source: arxiv.org. Saturation forecast: Around March 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 57.4 | #48 |
| 2 | Gemini 3.1 Pro (Preview) | 51.3 | #91 |
| 3 | GLM-5.3 | 51.3 | #62 |
| 4 | GPT-5.4 Mini | 42.6 | #389 |
| 5 | GPT-5.2 | 41.7 | #172 |
| 6 | GLM-5.2 | 33 | #137 |
| 7 | DeepSeek V4 Pro | 29.6 | #181 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=procarena-oracle-interactive-from-scratch-synthesis · How It Works · Data refreshed daily, snapshot 2026-11-12.