PRISM-VLM - Visual Discrimination: leaderboard
Metric: Per-axis mean score (%; forced A/B choice between the gold caption and a one-detail minimal edit (chance 50%); over PRISM-VLM's 6,238 items recycled from 15 public VLM benchmarks (five seeds of 100 items per benchmark); GPT-5 (low effort) synthesizes the perturbations and grades the open-ended axes; each model at the lowest reasoning effort its provider exposes, temperature 0). Source: arxiv.org. Saturation forecast: Around December 2026. 42 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (Minimal) | 93.7 |
| 2 | Gemini 2.5 Flash (Non-reasoning) | 91.5 |
| 3 | Gemini 2.0 Flash | 90 |
| 4 | Gemini 2.0 Flash Lite | 87.6 |
| 5 | Gemini 2.5 Flash Lite | 87.5 |
| 6 | GPT-4.1 Mini | 86.7 |
| 7 | Molmo2-8B | 86.4 |
| 8 | GPT-5 Mini (Minimal) | 86 |
| 9 | Qwen 3.5 9B (Non-reasoning) | 85.9 |
| 10 | Qwen 3 VL 8B Instruct | 85.8 |
| 11 | Gemma 4 E4B | 84 |
| 12 | Claude Haiku 4.5 | 83.9 |
| 13 | Qwen 3 VL 4B Instruct | 83 |
| 14 | GPT-5.4 Mini | 82.7 |
| 15 | Grok 4 Fast (Non-reasoning) | 82.5 |
Interactive version: theaggregate.ai/benchmark?slug=prism-vlm-visual-discrimination · How It Works · Data refreshed daily, snapshot 2026-09-26.