RuleWorld - Parallel Multi-Rule (Natural Language): leaderboard

Metric: Exact-match accuracy (%; x100 of the 0-1 score; parallel multi-rule QA with 2 to 8 gold rules, mean of the seven rule counts, rules 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 December 2026. 10 models tracked.

Top models

#ModelScore
1GPT-5.591.07
2Claude Sonnet 4.685.95
3GPT-5.176.02
4DeepSeek V3.2 (Non-reasoning)72.36
5Qwen 2.5 72B Instruct66.98
6Llama 3.1 70B Instruct56.33
7Qwen 2.5 32B Instruct53.57
8Qwen 3 32B49.33
9Llama 3.3 70B Instruct46.86
10Qwen 2.5 14B Instruct40.31

Interactive version: theaggregate.ai/benchmark?slug=ruleworld-parallel-multi-rule-natural-language · How It Works · Data refreshed daily, snapshot 2026-09-26.