EpiQAL - Text-Grounded Recall: leaderboard
EpiQAL-A (Emory University, University of Illinois Chicago and Microsoft, 2026): 475 multiple-choice questions whose correct options are stated verbatim in an open-access PLOS Neglected Tropical Diseases research article that the model reads in full, with distractors that match the wording but differ in role, population or context; a question may have several correct options. The score is exact match of the selected option set in percent, higher is better, zero-shot without chain-of-thought prompting, and models are told a question may have one, several or no correct options. Qwen3-30B-A3B-Instruct-2507 generated the items and GPT-5-mini, DeepSeek-V3.2-Thinking and GLM-4.5-Air verified the options.
Metric: Exact Match (%; zero-shot, no CoT, full article). Source: arxiv.org. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 90.5 | #349 |
| 2 | Mistral Large 2 (Nov) Instruct (2411) | 90.1 | #1199 |
| 3 | Qwen 3 30B A3B 2507 Instruct | 88.2 | #1078 |
| 4 | Qwen 3 32B | 87.2 | #960 |
| 5 | Qwen 3 8B | 81.1 | #1838 |
| 6 | Llama 3.1 8B Instruct | 79.8 | #4277 |
| 7 | Llama 3.3 70B Instruct [4bit] | 77.9 | #1221 |
| 8 | GPT-4.1 Nano | 76.8 | #1992 |
| 9 | GPT-4o Mini | 76.6 | #1627 |
| 10 | Mistral 7B Instruct (v0.3) | 72.2 | #6698 |
| 11 | Phi-4 Mini Instruct | 58.3 | #6452 |
| 12 | Llama 3.2 3B Instruct | 36.6 | #7185 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=epiqal-text-grounded-recall · How It Works · Data refreshed daily, snapshot 2026-11-12.