PRISM-VLM - Counting: leaderboard
Metric: Per-axis mean score (%; exact-integer accuracy on synthesized counting questions; 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 2030. 42 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 9B (Non-reasoning) | 66.4 |
| 2 | Gemini 3 Flash (Minimal) | 65.1 |
| 3 | Qwen 3.5 4B (Non-reasoning) | 64.6 |
| 4 | GPT-4.1 Mini | 61.8 |
| 5 | Qwen 3 VL 8B Instruct | 61.8 |
| 6 | GPT-5.4 Mini | 61.7 |
| 7 | Gemini 2.5 Flash (Non-reasoning) | 61.7 |
| 8 | GPT-5 Mini (Minimal) | 61.5 |
| 9 | Qwen 3 VL 4B Instruct | 58.5 |
| 10 | Qwen 3.5 2B (Non-reasoning) | 57.9 |
| 11 | Gemini 2.0 Flash | 57.3 |
| 12 | Nova 2 Lite | 56.2 |
| 13 | Gemini 2.0 Flash Lite | 55.1 |
| 14 | Gemma 4 E4B | 54.8 |
| 15 | Ministral-3-8B-Instruct-2512 | 54.5 |
Interactive version: theaggregate.ai/benchmark?slug=prism-vlm-counting · How It Works · Data refreshed daily, snapshot 2026-09-26.