EpiQAL - Multi-Step Inference: leaderboard

Metric: Exact Match (%; zero-shot, no CoT, full article). Source: arxiv.org. Saturation forecast: Around December 2026. 14 models tracked.

Top models

#ModelScore
1Mistral 7B Instruct (v0.3)78.9
2Qwen 3 32B74.3
3GPT-4.1 Nano67.8
4Mistral Large 2 (Nov) Instruct (2411)64.4
5Qwen 3 30B A3B 2507 Instruct57.3
6GPT-5 Mini53.3
7Qwen 3 8B50.8
8Llama 3.1 8B Instruct31.8
9Phi-4 Mini Instruct24.9
10GPT-4o Mini22.2
11Llama 3.2 3B Instruct15.7

Interactive version: theaggregate.ai/benchmark?slug=epiqal-multi-step-inference · How It Works · Data refreshed daily, snapshot 2026-09-25.