HELM VHELM - Seed Bench - Instance Interaction: leaderboard
VHELM (Stanford CRFM HELM for vision-language models, run v2.1.2) SEED-Bench instance interaction dimension: four-option multiple-choice image-understanding questions; 97 questions, prefix quasi-exact match. Percent correct, 0-100, higher is better; guessing scores 25.
Metric: Prefix Quasi-Exact Match (%). Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 48 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 (2025-04-16) | 86.6 | #125 |
| 2 | O1 (2024-12-17) | 86.6 | #157 |
| 3 | GPT-4.5 | 86.6 | #278 |
| 4 | GPT-4.1 Mini | 85.57 | #363 |
| 5 | Claude 3.5 Sonnet (20241022) | 84.54 | #302 |
| 6 | Gemini 2.5 Pro (Preview 03-25) | 84.54 | #126 |
| 7 | GPT-4 Turbo | 84.54 | #417 |
| 8 | Gemini 1.5 Flash (002) | 83.51 | #587 |
| 9 | GPT-4o (2024-05-13) | 82.47 | #337 |
| 10 | GPT-4o (2024-11-20) | 82.47 | #392 |
| 11 | O4 Mini (2025-04-16) | 81.44 | #183 |
| 12 | GPT-4.1 (2025-04-14) | 80.41 | #257 |
| 13 | GPT-4o (2024-08-06) | 80.41 | #343 |
| 14 | Gemini 2.0 Flash (Preview) | 80.41 | #376 |
| 15 | Gemini 1.5 Pro (002) | 79.38 | #368 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-seed-bench-instance-interaction · How It Works · Data refreshed daily, snapshot 2026-11-05.