MRI-Eval - Stem-Only: leaderboard
MRI-Eval (2026) tests MRI physics and GE scanner operations knowledge on 1,365 items drawn from textbooks, GE manuals, course material and expert questions. In the stem-only condition the answer options are removed and a Grok judge marks each free-text answer against the key; accuracy in percent, higher is better.
Metric: Stem-only accuracy (%): the answer options are removed and the free-text answer is marked correct or incorrect by a Grok judge (grok-4-1-fast-non-reasoning) against the keyed answer, over all 1,365 MRI-Eval items (MRI physics and GE scanner operations; GE questions carry a GE SIGNA Premier 3T context line), zero-shot at temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 61.1 | #113 |
| 2 | GPT-5.4 | 60.3 | #135 |
| 3 | Claude Sonnet 4.6 | 58.4 | #173 |
| 4 | Llama 3.3 70B Instruct | 37.1 | #1385 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=mri-eval-stem-only · How It Works · Data refreshed daily, snapshot 2026-11-21.