RuleWorld - Parallel Multi-Rule (First-Order Logic): 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 first-order logic; 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
1Claude Sonnet 4.691.17
2GPT-5.585.93
3DeepSeek V3.2 (Non-reasoning)78.21
4GPT-5.175.76
5Qwen 2.5 72B Instruct59.45
6Llama 3.1 70B Instruct55.71
7Qwen 3 32B53.26
8Qwen 2.5 32B Instruct50.38
9Llama 3.3 70B Instruct45.55
10Qwen 2.5 14B Instruct44.86

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