Visual OPST - POPE F1: leaderboard
OPST v1: Gemma 4 E2B/E4B Base means instruction-tuned; all training methods are separate checkpoints. OPST evaluates EMA teacher, not student. Nonthinking multiple-choice uses answer-option log-likelihood (MMLU 5-shot, other MC zero-shot, length normalization per lm-eval default); thinking parses final Answer: X. GSM8K 8-shot CoT exact numerical match; MATH-500 symbolic equivalence. Training topics and supervision preserved; no unpublished checkpoint URL or decoding settings inferred. Visual training uses unlabeled OpenImages V4; caption/spatial supervision are separate trained variants. POPE F1. Metric: Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | OPST study Gemma 4 E4B Visual OPST caption=× \times spatial=✓ \checkmark | 87.9 | |
| 2 | OPST study Gemma 4 E4B Visual OPST caption=✓ \checkmark spatial=× \times | 87.8 | |
| 3 | OPST study Gemma 4 E4B Visual OPST caption=✓ \checkmark spatial=✓ \checkmark | 87 | |
| 4 | OPST study Gemma 4 E4B Base | 84.8 | |
| 5 | OPST study Gemma 4 E2B Visual OPST caption=× \times spatial=✓ \checkmark | 84.7 | |
| 6 | OPST study Gemma 4 E2B Visual OPST caption=✓ \checkmark spatial=✓ \checkmark | 84.6 | |
| 7 | OPST study Gemma 4 E2B Visual OPST caption=✓ \checkmark spatial=× \times | 83 | |
| 8 | OPST study Gemma 4 E2B Base | 82.2 |
Interactive version: theaggregate.ai/benchmark?slug=visual-opst-pope-f1 · How It Works · Data refreshed daily, snapshot 2026-10-27.