MRI-Eval: leaderboard
MRI-Eval is a text-only benchmark for relative comparison of models on MRI physics and GE scanner operations knowledge. It evaluates primary multiple-choice questions, with additional stem-only and primed stem-only analyses. The pinned metric is accuracy, measuring performance on the scored items; the source notes raw outputs should be treated cautiously for GE-specific protocol guidance. Source: BenchmarkList’s attributed result table.
Metric: Accuracy (%). Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 97.1 | #83 |
| 2 | Claude Opus 4.6 | 96.6 | #67 |
| 3 | Claude Sonnet 4.6 | 95.2 | #94 |
| 4 | Gemini 2.5 Pro | 94.7 | #159 |
| 5 | Llama 3.3 70B | 93.2 | #578 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mri-eval · How It Works · Data refreshed daily, snapshot 2026-11-05.