Refresh Models —Deals —Regulatory —Sources —

InternVL-3-78B vs Qwen3 VL 4B (Reasoning)

11 SHARED BENCHMARKS

Across 11 shared benchmarks, InternVL-3-78B scores higher on 3 and Qwen3 VL 4B (Reasoning) on 8. The widest gap is mathvision, where Qwen3 VL 4B (Reasoning) scores 60 against 34.8.

OPENGVLABVSALIBABA11 SHARED3–8 HEAD-TO-HEAD

BenchmarkInternVL-3-78BMARGINQwen3 VL 4B (Reasoning)

AI2D83.584.9

BLINK51.965.1

ChartQAPro44.436.2

DocVQA-val83.894.7

HallusionBench40.264.1

mathvision34.860

MathVista70.179.5

MMMU (val) (Pass@1)60.870.8

MMStar66.173.2

RealWorldQA74.373.2

TextVQA-val83.580.5

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 4B (Reasoning) withEVERY TRACKED PAIRING

▶Anthropic12 PAIRINGS

Claude Opus 4.8 Claude Opus 4.6 Claude Sonnet 4.5 Claude 3.5 Sonnet Claude Fable 5 Claude Opus 4.5 Claude Opus 5 Claude Opus 4.7 Claude Sonnet 4 Claude Sonnet 4.6 Claude Sonnet 5 Claude Opus 4

▶OpenAI11 PAIRINGS

GPT-5.5 GPT-4o GPT-5.6 Sol GPT-5.2 GPT-5.4 GPT-5 o3 GPT-5.1 o4-mini o1 GPT-4o mini

▶Google11 PAIRINGS

Gemini 3.1 Pro Gemini 3 Pro Gemma 4 26B A4B Gemini 2.5 Pro Gemini 1.5 Pro Gemma 4 31B Gemini 3.5 Flash Gemini 3 Flash Preview Gemini 3.6 Flash Gemma 4 E2B Gemini 2.5 Flash

▶Alibaba10 PAIRINGS

Qwen3.5 35B A3B Qwen3.5 397B A17B Qwen2.5 Instruct 72B Qwen3.5 27B Qwen3.6 Plus Qwen3.6 27B Qwen3.6 35B A3B Qwen3.7 Max Qwen3.5 122B A10B Qwen3.7 Plus Preview

▶DeepSeek4 PAIRINGS

DeepSeek-V3 DeepSeek-V4-Pro DeepSeek-V3.2 DeepSeek-V4-Flash

▶Moonshot4 PAIRINGS

Kimi K2.5 Kimi K2.6 Kimi K3 Kimi K2 Instruct

▶Meta2 PAIRINGS

Llama 3.1 Instruct 405B Muse Spark 1.1

▶Z.ai2 PAIRINGS

GLM-5.1 GLM-5.2

▶MiniMax1 PAIRING

▶NVIDIA2 PAIRINGS

Nemotron 3 Super 120B A12B Nemotron 3 Ultra 550B A55B