A2RBench - Symbolic Tasks (Original Symbols): leaderboard

Symbolic, structure-based rule tasks with their original symbols, from A2RBench (2026), an automatically generated abstract reasoning benchmark in which a model infers a bijective transformation rule over sequences, grids or voxels from examples and applies it to a query; answers are judged against code-executed ground truth. Accuracy in percent; higher is better.

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 March 2028. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-541.3#178
2GPT-5 Mini40.2#364
3Gemini 3 Pro39.3#145
4O4 Mini36.5#368
5Claude Sonnet 4.534.2#261
6GPT-5.232.5#185
7Gemini 2.5 Flash29.3#463
8GLM-4.624.8#498
9GPT-4o Mini23.9#1667
10DeepSeek V3.223.4#419
11Qwen 3 32B23.1#972
12Qwen 3 14B16.8#1269
13Gemini 3 Flash15.7#175

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=a2rbench-symbolic-tasks-original-symbols · How It Works · Data refreshed daily, snapshot 2026-11-21.