Which is better, Qwen3.5 4B or Qwen3 VL 4B (Reasoning)?

Qwen3.5 4B vs Qwen3 VL 4B (Reasoning)

31 SHARED BENCHMARKS

Across 31 shared benchmarks, Qwen3.5 4B scores higher on 26 and Qwen3 VL 4B (Reasoning) on 4, with 1 level. The widest gap is τ²-Bench Telecom (AA run), where Qwen3.5 4B scores 92.1 against 23.4.

ALIBABAVSALIBABA31 SHARED26–4 HEAD-TO-HEAD

BenchmarkQwen3.5 4BMARGINQwen3 VL 4B (Reasoning)

AA Agentic Index36.314.4

AA Intelligence207

AA-LCR6321.3

AA-Omniscience-58.4-68.4

Artificial Analysis Coding Index22.66.7

BFCL v371.167.3

critpt00

ERQA46.347.3

gdpval8.313.8

GPQA Diamond77.164.1

HLE9.94.6

HMMT 202576.853.1

IFBench59.236.6

ifeval89.882.6

include7164.6

LiveCodeBench v655.851.3

MathVista63.679.5

mmlu_prox71.565

mmlu_redux88.886

MMLU-Pro79.173.6

MMMU (val) (Pass@1)50.370.8

MMMU-Pro65.457

MMStar75.373.2

OCRBench86.981.6

OmniScience Accuracy15.112.2

OmniScience Non-Hallucination13.48.2

RealWorldQA76.373.2

scicode18.317.1

supergpqa52.946.8

Terminal-Bench Hard18.21.5

τ²-Bench Telecom (AA run)92.123.4

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 Qwen3.5 4B withALL PAIRINGS →

AlibabaQwen3.5 35B A3B Qwen3.5 397B A17B Qwen2.5 Instruct 72B Qwen3.5 27B Qwen3.6 Plus Qwen3.6 35B A3B

Across the fieldGemma 4 26B A4B Kimi K3 Llama 3.1 Instruct 405B DeepSeek-V3.2 DeepSeek-V4-Flash Claude Opus 4.7 Gemini 2.5 Pro Gemini 1.5 Pro

Compare Qwen3 VL 4B (Reasoning) withALL PAIRINGS →

AlibabaQwen3.5 35B A3B Qwen3.5 397B A17B Qwen2.5 Instruct 72B Qwen3.5 27B Qwen3.6 Plus Qwen3.6 35B A3B

Across the fieldGemini 3.6 Flash Nemotron 3 Super 120B A12B GPT-5.5 Claude Opus 4.8 GPT-4o GPT-5.6 Sol Gemini 3.1 Pro GPT-5.4