← All models

Qwen3 VL 2B Thinking

Alibaba · 2025-10-21 · 2.1B parameters

Qwen3-VL-2B-Thinking is the reasoning-enabled variant of the smallest (2B dense) model in Alibaba's Qwen3-VL vision-language family, released October 21, 2025 under Apache 2.0 together with Qwen3-VL-32B. It always produces a thinking trace before answering and keeps the family's image and video understanding, OCR, grounding and visual-agent features. Its context is 262,144 tokens natively and expandable to 1 million tokens.

Benchmark scores