ECG-Reasoning-Benchmark (MIMIC-IV-ECG) - Initial Diagnosis Accuracy: leaderboard
Metric: Initial diagnosis accuracy (%): yes/no answer to the initial diagnostic question before any step-wise reasoning (chance 50), on the 3,355 MIMIC-IV-ECG samples, multi-turn ECG reasoning chains for 17 core diagnoses (balanced positive and negative cases), 12-lead ECGs given as plotted images (OpenTSLM as a 100 Hz time series), temperature 0; each step's answer is checked for semantic agreement with the ground truth by a Gemini-3-Flash judge; higher is better. Source: arxiv.org. Saturation forecast: Around 2031. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 63.41 | #105 |
| 2 | Gemini 3 Flash (Preview) | 60.64 | #78 |
| 3 | Gemini 2.5 Pro | 54.69 | #145 |
| 4 | Qwen 3 VL 32B Instruct | 54.3 | #276 |
| 5 | Gemini 2.5 Flash | 53.74 | #237 |
| 6 | Qwen 3 VL 8B Instruct | 53.53 | #401 |
| 7 | Llama 3.2 90B Vision Instruct | 52.52 | #611 |
| 8 | GPT-5 Mini (2025-08-07) | 51.5 | #165 |
| 9 | Llama 3.2 11B Instruct | 49.75 | #1112 |
| 10 | MedGemma-4B-IT | 49.3 | #842 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=ecg-reasoning-benchmark-mimic-iv-ecg-initial-diagnosis-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-11.