RuleWorld - Multi-Hop (Natural Language): leaderboard

Metric: Exact-match accuracy (%; x100 of the 0-1 score; multi-hop rule QA with 2 to 4 hops, mean of the three hop 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 March 2027. 10 models tracked.

Top models

#ModelScore
1Claude Sonnet 4.664.33
2GPT-5.546.67
3DeepSeek V3.2 (Non-reasoning)38
4Qwen 3 32B33.33
5GPT-5.128
6Qwen 2.5 32B Instruct22
7Qwen 2.5 14B Instruct19.33
8Llama 3.3 70B Instruct18.67
9Qwen 2.5 72B Instruct18.67
10Llama 3.1 70B Instruct14.67

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