PRISM-VLM - Multi-Question Robustness: leaderboard
The multi-question axis of PRISM-VLM (NAVER, 2026): three to five questions about the same image are asked in one prompt, and an item counts only if every answer is correct. Share of items answered fully correctly, in percent; higher is better.
Metric: Per-axis mean score (%; share of items whose 3-5 bundled questions about one image are all answered correctly; 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 June 2027. 42 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Minimal) | 56 | #159 (Gemini 3 Flash) |
| 2 | Gemini 3.1 Flash Lite (Minimal) | 53 | #363 (Gemini 3.1 Flash Lite) |
| 3 | GPT-4.1 Mini | 50.7 | #775 |
| 4 | Qwen 3.5 9B (Non-reasoning) | 50.5 | #803 (Qwen 3.5 9B) |
| 5 | GPT-5 Mini (Minimal) | 49.4 | #349 (GPT-5 Mini) |
| 6 | Gemini 2.5 Flash (Non-reasoning) | 48.7 | #452 (Gemini 2.5 Flash) |
| 7 | GPT-5.4 Mini | 48.6 | #389 |
| 8 | Qwen 3.5 4B (Non-reasoning) | 48.5 | #1180 (Qwen 3.5 4B) |
| 9 | Gemini 2.0 Flash | 46.8 | #702 |
| 10 | Qwen 3 VL 8B Instruct | 44.4 | #982 |
| 11 | Gemini 2.0 Flash Lite | 43.7 | #1032 |
| 12 | Qwen 3 VL 4B Instruct | 41.5 | #1281 |
| 13 | Gemini 2.5 Flash Lite | 40.8 | #958 |
| 14 | Molmo2-8B | 39.7 | #1488 |
| 15 | Nova 2.0 Lite | 38.7 | #1133 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=prism-vlm-multi-question-robustness · How It Works · Data refreshed daily, snapshot 2026-11-12.