Atlas

Models

← All models

Qwen3 VL 8B Thinking

Alibaba · 2025-10-15 · 8.8B parameters

Qwen3-VL-8B-Thinking is the reasoning-enabled variant of Alibaba's 8B Qwen3-VL vision-language model. It provides multimodal understanding, OCR, grounding, and visual-agent capabilities in a smaller model with thinking enabled. Its context is 262,144 tokens natively and expandable to 1 million tokens.

Benchmark scores

BenchmarkScore
AA-LCR31.0
AA-LCR v1.133.3
AA-Omniscience Index-52.3
AIME 2025 (Artificial Analysis avg@10)30.7
Artificial Analysis Omniscience Accuracy20.4
Artificial Analysis Omniscience Hallucination Rate91.4
Artificial Analysis Openness Index50.0
Artificial Analysis Progress Index v1185
Capability131.4
CritPt (avg@5)0.3
FormationEval90.3
GPQA Diamond (Artificial Analysis avg@5)57.9
Humanity's Last Exam (Text-Only)3.8
IFBench (Artificial Analysis avg@5)39.9
LiveCodeBench (Artificial Analysis)35.3
MATH-Vision59.6
MMLU Pro74.9
MMMU Pro56.6
ObviousBench86.1
SciCode (Artificial Analysis avg@3)21.9
SnakeBench Average Apples2.9
SnakeBench Best Apples7.0
SnakeBench Rating20.5
SnakeBench Total Apples57.0
SnakeBench Win Rate33.3
Terminal-Bench Hard3.8
τ²-bench (Telecom)22.5
Loading Atlas data…