RuleWorld: leaderboard
RuleWorld (2026) tests whether LLMs can locate and apply externally provided procedural rules from a large pool of abstract, deliberately non-commonsense rules. With all 1,000 rules of the pool in the prompt, the score is the mean of six exact-match accuracies (single-rule, parallel multi-rule and multi-hop QA, with rules in first-order logic and in natural language), in percent; higher is better.
Metric: Exact-match accuracy (%; x100 of the 0-1 score; mean of the six task-representation cells (single-rule, parallel multi-rule over 2-8 rules, and 2-4 hop multi-hop QA, each with rules in first-order logic and in natural language); all 1,000 abstract, non-commonsense procedural rules of the pool prepended to the prompt (full-context prompting); exact match on the boxed answer list, partial credit per matching position; 10 questions per task type and difficulty level for each of five seeds; temperature 0 (Qwen3-32B 0.6)). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 77.63 | #158 |
| 2 | GPT-5.5 (2026-04-23) | 72.83 | #33 |
| 3 | GPT-5.1 (2025-11-13) | 67.96 | #184 |
| 4 | DeepSeek V3.2 (Non-reasoning) | 66.87 | #403 (DeepSeek V3.2) |
| 5 | Llama 3.1 70B Instruct | 54.9 | #1541 |
| 6 | Qwen 2.5 72B Instruct | 53.07 | #1093 |
| 7 | Llama 3.3 70B Instruct | 51.62 | #1353 |
| 8 | Qwen 3 32B | 51.32 | #960 |
| 9 | Qwen 2.5 32B Instruct | 50.77 | #1216 |
| 10 | Qwen 2.5 14B Instruct | 42.97 | #1745 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ruleworld · How It Works · Data refreshed daily, snapshot 2026-11-12.