Atlas

Models

← All models

Qwen2.5 VL 32B Instruct

Alibaba · 2025-03-24 · 33.5B parameters

Qwen2.5-VL-32B-Instruct is Alibaba's instruction-tuned vision-language model released in March 2025 under Apache 2.0. It supports dynamic image resolutions, document parsing, OCR, visual reasoning, screenshot-based code understanding, and video analysis within its 128K-token context window.

Benchmark scores

BenchmarkScore
Arena Vision — English (No Style Control)1159
Arena Vision — English (Style Controlled)1124
OSWorld-Verified3.9
ScreenSpot-Pro48.0
SnakeBench Average Apples1.8
SnakeBench Best Apples5.0
SnakeBench Rating16.7
SnakeBench Total Apples60.0
SnakeBench Win Rate44.4
Vision Arena Overall1119
Vision Arena Overall No Style Control1152
Loading Atlas data…