PRISM-VLM: leaderboard

Metric: PScore (%; harmonic mean of the seven per-axis mean scores, x100, 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

#ModelScore
1Gemini 3 Flash (Minimal)71.9
2Gemini 2.5 Flash (Non-reasoning)67.3
3Qwen 3 VL 8B Instruct67.1
4Qwen 3 VL 4B Instruct64.2
5GPT-5 Mini (Minimal)63.3
6Nova 2 Lite62.8
7Gemini 2.5 Flash Lite62.1
8Qwen 3.5 9B (Non-reasoning)61.1
9Qwen 3.5 4B (Non-reasoning)59.8
10Gemini 2.0 Flash Lite59.1
11GPT-4.1 Mini57.6
12Gemini 2.0 Flash57
13Grok 4 Fast (Non-reasoning)55.6
14Qwen 3.5 2B (Non-reasoning)54.9
15Ministral-3-8B-Instruct-251253.7

Interactive version: theaggregate.ai/benchmark?slug=prism-vlm · How It Works · Data refreshed daily, snapshot 2026-09-26.