EpiQAL - Masked-Input Reasoning: 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)80.8
2Mistral Large 2 (Nov) Instruct (2411)68.5
3GPT-5 Mini59.9
4Qwen 3 30B A3B 2507 Instruct58.5
5GPT-4.1 Nano55.9
6Qwen 3 32B54.7
7Qwen 3 8B48.4
8Phi-4 Mini Instruct42.6
9GPT-4o Mini21.3
10Llama 3.1 8B Instruct19
11Llama 3.2 3B Instruct8.8

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