LookZoom Pointer Comparison Replace 98 - Five-Benchmark Average: leaderboard
LookZoom v1: Qwen3-VL-Instruct at 2B/4B/8B. Fixed 500-question subsets (seed 0) of DocVQA val (ANLS),InfographicVQA val (ANLS),ChartQA test (relaxed accuracy),TextVQA val (VQA accuracy), plus 300 held-out V-CoT-Doc (strict match). Per-view budget B means B * 32^2 pixels preserving aspect ratio, not total tokens. Single view/AttWarp/replace answer from 1 view; ViCrop/ground-then-crop/GapSight/LookZoom use 2. LookZoom reader uses GRPO LoRA; pointer trained from own correct answers; GapSight is authors reimplementation. Distinct trained systems remain separate. Appendix S12: identical adaptive window for two-view comparison; replace uses only zoomed view. Pointer and view retained in identity. Five-Benchmark Average. Metric: Accuracy (%).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | LookZoom Qwen3-VL-4B Adaptive window, <look> token [Replace] | 61.2 | |
| 2 | LookZoom Qwen3-VL-4B Square .5 .5 , <look> token [Replace] | 55.1 | |
| 3 | LookZoom Qwen3-VL-4B Square .5 .5 , pre-answer [Replace] | 51.9 | |
| 4 | LookZoom Qwen3-VL-4B Adaptive window, pre-answer [Replace] | 51.8 | |
| 5 | LookZoom Qwen3-VL-4B Fovea, <look> token [Replace] | 49.8 | |
| 6 | LookZoom Qwen3-VL-4B Fovea, pre-answer [Replace] | 42.7 |
Interactive version: theaggregate.ai/benchmark?slug=lookzoom-pointer-comparison-replace-98-five-benchmark-average · How It Works · Data refreshed daily, snapshot 2026-10-27.