Request a pilot
‹ PublicAI Index

The LLM benchmark aggregator.

Qwen3 VL 8B Instruct

Alibaba · 8.8B · open weights

Strongest in Document parsing (#41 of 76), weakest in Knowledge (#83 of 190). Above par in 2 of 2 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5.5 and Claude Fable 5.1 and ahead of Gemini 3.5 Flash Lite and Claude Haiku 5.5.

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Knowledge52.1#83/190
Document parsing52.3#41/76

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Qwen3 VL 8B Instruct, left for the other.

GLM OCRwon 2 · lost 0
Qwen3.5 0.8Bwon 2 · lost 0
WeVisDoc 4Bwon 2 · lost 0
Qwen3.5 2Bwon 2 · lost 0
Granite Vision 4.1 4Bwon 2 · lost 0
WeVisDoc 2Bwon 2 · lost 0
Qwen3.5 9Bwon 2 · lost 0
Nova 2 Litewon 2 · lost 0
MinerU Diffusion V1 2.5Bwon 2 · lost 0
Gemma 4 E4Bwon 2 · lost 0
DeepSeek OCR 2won 2 · lost 0
PaddleOCR VLwon 2 · lost 0
Qwen3.5 4Bwon 2 · lost 0
GPT-5.4 Nanowon 2 · lost 0
OvisOCR2won 2 · lost 0
MinerU2.5 1.2Bwon 2 · lost 0
Qianfan OCRwon 2 · lost 0
Docling Modelswon 2 · lost 0
Nemotron 3 Nano Omni 30B A3Bwon 2 · lost 0
Unlimited OCRwon 2 · lost 0
Jina OCR V1won 2 · lost 0
Qwen3.6 35B A3Bwon 2 · lost 0
Falcon OCRwon 2 · lost 0
LightOnOCR 2 1Bwon 2 · lost 0
HPD Parsingwon 2 · lost 0
GPT-5 Miniwon 2 · lost 0
Claude Haiku 4.5won 2 · lost 0
TeleOCRwon 2 · lost 0
Gemma 4 26B A4Bwon 2 · lost 0
Dots.mocrwon 2 · lost 0
Gemini 3.1 Flash Litewon 2 · lost 0
Claude Opus 4.6won 2 · lost 0
Claude Haiku 5.5won 2 · lost 0
Gemini 3.5 Flash Litewon 2 · lost 0
GPT-5.4won 1 · lost 1
HunyuanOCR 1.5won 0 · lost 2
PaddleOCR VL 1.5won 0 · lost 2
PaddleOCR VL 1.6won 0 · lost 2
Gemma 4 31Bwon 0 · lost 2
Claude Sonnet 5won 0 · lost 2
Surya OCR 2won 0 · lost 2
GPT-6 Lunawon 0 · lost 2
DeepSeek V4.1 Flashwon 0 · lost 2
Claude Opus 4.8won 0 · lost 2
Claude Opus 4.7won 0 · lost 2
Chandra Ocr 2won 0 · lost 2
GPT-5.6 Lunawon 0 · lost 2
GPT-5.5won 0 · lost 2
GPT-5.6 Terrawon 0 · lost 2
Qwen3.8 27Bwon 0 · lost 2
GPT-6 Solwon 0 · lost 2
KDL Frontier Parser Nanowon 0 · lost 2
LightOnOCR 3 1Bwon 0 · lost 2
Gemini 3.1 Prowon 0 · lost 2
GPT-6.1 Solwon 0 · lost 2
Gemini 3.6 Flashwon 0 · lost 2
MinerU2.5 Pro 1.2Bwon 0 · lost 2
GLM-5.3 Flashwon 0 · lost 2
GPT-6 Astrawon 0 · lost 2
Claude Sonnet 5.5won 0 · lost 2
Infinity Parser2 Flashwon 0 · lost 2
Gemini 3.7 Flashwon 0 · lost 2
Claude Fable 5won 0 · lost 2
Infinity Parser2 Prowon 0 · lost 2
Gemini 3.8 Flashwon 0 · lost 2
LightOnOCR 3 0.8Bwon 0 · lost 2
Gemini 3.5 Flashwon 0 · lost 2
LightOnOCR 3 4Bwon 0 · lost 2
Florin Parser Nanowon 0 · lost 2
aeddix-alpine-ocr (KDL pipeline)won 0 · lost 2
GPT-5.6 Solwon 0 · lost 2
Gemini 3 Flashwon 0 · lost 2
Rakedoc Nanowon 0 · lost 2
Claude Fable 5.1won 0 · lost 2
Claude Opus 5.5won 0 · lost 2

§ 3 · Sources

Where the numbers come from

1 publication, 6 figures. Every one links to the page it was read from.

ParseBench ↗6 measuresread 2026-10-19
ParseBench · Overall 61.97%ParseBench · Tables 74.61%ParseBench · Charts 28.18%ParseBench · Content faithfulness 87.63%ParseBench · Semantic formatting 64.23%ParseBench · Visual grounding 55.18%

Badge

PublicAI Index badge for Qwen3 VL 8B Instruct[![PublicAI Index](https://publicai.io/model-index/badge?model=qwen3-vl-8b-instruct)](https://publicai.io/model-index/m/qwen3-vl-8b-instruct)