VECTOR WIREAI INTELLIGENCE
⌘K
UTC
Refresh Models —Deals —Regulatory —Sources —

Qwen3 VL 235B A22B Reasoning vs Qwen3 VL 4B (Reasoning)

52 SHARED BENCHMARKS

Across 52 shared benchmarks, Qwen3 VL 235B A22B Reasoning scores higher on 50 and Qwen3 VL 4B (Reasoning) on 1, with 1 level. The widest gap is Terminal-Bench Hard, where Qwen3 VL 235B A22B Reasoning scores 11.4 against 1.5.

ALIBABAVSALIBABA52 SHARED50–1 HEAD-TO-HEAD
BenchmarkQwen3 VL 235B A22B ReasoningMARGINQwen3 VL 4B (Reasoning)
AA Intelligence20.97.7
AA-LCR6323
AA-Omniscience-46.5-68.4
AI2D89.284.9
AIME 202589.774.5
BFCL v371.967.3
BLINK67.165.1
CC-OCR81.573.8
CharadesSTA63.559
charxiv_rq66.148.5
critpt00
DocVQA_test96.594.2
EmbSpatial-Bench84.380.7
ERQA52.547.3
GPQA Diamond77.264.1
HallusionBench66.764.1
HLE13.64.6
HMMT 202577.453.1
IFBench56.536.6
ifeval88.282.6
include8064.6
InfoVQA (val)89.579.5
InfoVQA_test89.583
LiveCodeBench69.551.3
LiveCodeBench v670.151.3
LongVideoBench65.657.7
LVBench63.653.5
mathvision74.660
MathVista85.879.5
MM MTBench8.57.7
mmlu_prox80.665
mmlu_redux93.786
MMLU-Pro83.873.6
MMMU (val) (Pass@1)80.670.8
MMMU-Pro69.357
MMStar78.773.2
OCRBench87.581.6
OCRBench v2_en66.861.8
OmniScience Accuracy20.912.2
OmniScience Non-Hallucination14.88.2
OSWorld-Verified38.131.4
RealWorldQA81.373.2
RefSpatial-Bench0.745.3
RoboSpatial-Home73.963.2
scicode39.917.1
ScreenSpot95.492.9
screenspot_pro_no_tools61.849.2
supergpqa64.346.8
Terminal-Bench Hard11.41.5
Video-MME7959.7
VideoMMMU8069.4
τ²-Bench Telecom (AA run)54.115.5

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.

Compare Qwen3 VL 235B A22B Reasoning withEVERY TRACKED PAIRING

▶Anthropic12 PAIRINGS
Claude Opus 4.8Claude Sonnet 4.5Claude Opus 4.6Claude 3.5 SonnetClaude Fable 5Claude Opus 4.5Claude Opus 5Claude Opus 4.7Claude Sonnet 4Claude Sonnet 4.6Claude Sonnet 5Claude Opus 4
▶OpenAI11 PAIRINGS
GPT-5.5GPT-4oGPT-5.6 SolGPT-5.4GPT-5.2GPT-5o3GPT-5.1o4-minio1GPT-4o mini
▶Google10 PAIRINGS
Gemini 3.1 ProGemini 3 ProGemma 4 26B A4BGemini 2.5 ProGemini 1.5 ProGemma 4 31BGemini 3.5 FlashGemini 3.6 FlashGemini 3 Flash PreviewGemma 4 E2B
▶Alibaba10 PAIRINGS
Qwen3.5 35B A3BQwen2.5 Instruct 72BQwen3.5 397B A17BQwen3.5 27BQwen3.6 PlusQwen3.6 35B A3BQwen3.6 27BQwen3.7 MaxQwen3.7 Plus PreviewQwen3.5 122B A10B
▶DeepSeek4 PAIRINGS
DeepSeek-V3DeepSeek-V4-ProDeepSeek-V3.2DeepSeek-V4-Flash
▶Moonshot4 PAIRINGS
Kimi K2.5Kimi K2.6Kimi K3Kimi K2 Instruct
▶Meta2 PAIRINGS
Llama 3.1 Instruct 405BMuse Spark 1.1
▶Z.ai2 PAIRINGS
GLM-5.1GLM-5.2
▶MiniMax1 PAIRING
MiniMax M2.7
▶NVIDIA2 PAIRINGS
Nemotron 3 Super 120B A12BNemotron 3 Ultra 550B A55B

Compare Qwen3 VL 4B (Reasoning) withEVERY TRACKED PAIRING

▶Anthropic12 PAIRINGS
Claude Opus 4.8Claude Sonnet 4.5Claude Opus 4.6Claude 3.5 SonnetClaude Fable 5Claude Opus 4.5Claude Opus 5Claude Opus 4.7Claude Sonnet 4Claude Sonnet 4.6Claude Sonnet 5Claude Opus 4
▶OpenAI11 PAIRINGS
GPT-5.5GPT-4oGPT-5.6 SolGPT-5.4GPT-5.2GPT-5o3GPT-5.1o4-minio1GPT-4o mini
▶Google10 PAIRINGS
Gemini 3.1 ProGemini 3 ProGemma 4 26B A4BGemini 2.5 ProGemini 1.5 ProGemma 4 31BGemini 3.5 FlashGemini 3.6 FlashGemini 3 Flash PreviewGemma 4 E2B
▶Alibaba10 PAIRINGS
Qwen3.5 35B A3BQwen2.5 Instruct 72BQwen3.5 397B A17BQwen3.5 27BQwen3.6 PlusQwen3.6 35B A3BQwen3.6 27BQwen3.7 MaxQwen3.7 Plus PreviewQwen3.5 122B A10B
▶DeepSeek4 PAIRINGS
DeepSeek-V3DeepSeek-V4-ProDeepSeek-V3.2DeepSeek-V4-Flash
▶Moonshot4 PAIRINGS
Kimi K2.5Kimi K2.6Kimi K3Kimi K2 Instruct
▶Meta2 PAIRINGS
Llama 3.1 Instruct 405BMuse Spark 1.1
▶Z.ai2 PAIRINGS
GLM-5.1GLM-5.2
▶MiniMax1 PAIRING
MiniMax M2.7
▶NVIDIA2 PAIRINGS
Nemotron 3 Super 120B A12BNemotron 3 Ultra 550B A55B