A2RBench - Symbolic Tasks (Original Symbols): leaderboard
Metric: Accuracy (%) on the 351 symbolic (structure-based) rule tasks with their original symbols (P0), the solver states its inferred rule and reasoning, then its answer at near-zero temperature (1e-7) is judged against the code-executed ground truth by a GPT-5-mini judge; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 41.3 |
| 2 | GPT-5 Mini | 40.2 |
| 3 | Gemini 3 Pro | 39.3 |
| 4 | O4 Mini | 36.5 |
| 5 | Claude Sonnet 4.5 | 34.2 |
| 6 | GPT-5.2 | 32.5 |
| 7 | Gemini 2.5 Flash | 29.3 |
| 8 | GLM-4.6 | 24.8 |
| 9 | GPT-4o Mini | 23.9 |
| 10 | DeepSeek V3.2 | 23.4 |
| 11 | Qwen 3 32B | 23.1 |
| 12 | Qwen 3 14B | 16.8 |
| 13 | Gemini 3 Flash | 15.7 |
Interactive version: theaggregate.ai/benchmark?slug=a2rbench-symbolic-tasks-original-symbols · How It Works · Data refreshed daily, snapshot 2026-10-07.