A2RBench - Semantic Tasks: leaderboard
Metric: Accuracy (%) on the 352 semantic (knowledge-dependent) rule tasks, 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 January 2027. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 52 |
| 2 | Gemini 3 Pro | 48.6 |
| 3 | GPT-5 Mini | 47.7 |
| 4 | O4 Mini | 44.3 |
| 5 | Gemini 2.5 Flash | 30.4 |
| 6 | Claude Sonnet 4.5 | 25.3 |
| 7 | GPT-5.2 | 23.9 |
| 8 | DeepSeek V3.2 | 20.7 |
| 9 | GLM-4.6 | 16.8 |
| 10 | Gemini 3 Flash | 13.4 |
| 11 | Qwen 3 32B | 12.8 |
| 12 | GPT-4o Mini | 9.4 |
| 13 | Qwen 3 14B | 8.8 |
Interactive version: theaggregate.ai/benchmark?slug=a2rbench-semantic-tasks · How It Works · Data refreshed daily, snapshot 2026-10-07.