HELM VHELM - Mmmu - Architecture and Engineering: leaderboard
Metric: Prefix Quasi-Exact Match (%). Source: crfm.stanford.edu. 48 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O3 (2025-04-16) | 89.66 |
| 2 | O4 Mini (2025-04-16) | 79.31 |
| 3 | O1 (2024-12-17) | 65.52 |
| 4 | GPT-4o Mini (2024-07-18) | 55.17 |
| 5 | GPT-4.1 (2025-04-14) | 55.17 |
| 6 | GPT-4o (2024-05-13) | 55.17 |
| 7 | Claude 3.5 Sonnet (20240620) | 51.72 |
| 8 | Gemini 2.5 Pro (Preview 03-25) | 51.72 |
| 9 | Gemini 2.0 Flash (Preview) | 51.72 |
| 10 | Gemini 2.0 Flash | 48.28 |
| 11 | Gemini 2.0 Pro (Preview 02-05) | 48.28 |
| 12 | Gemini 2.0 Flash Lite | 48.28 |
| 13 | Gemini 1.5 Pro (002) | 44.83 |
| 14 | GPT-4o (2024-08-06) | 44.83 |
| 15 | Claude 3 Opus (20240229) | 44.83 |
Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-mmmu-architecture-and-engineering · How It Works · Data refreshed daily, snapshot 2026-09-19.