Captioning Benchmark
Norges Bank Report (132 pages)
| Parsing | Captioning | Accelerator | Tokens | Seconds |
|---|---|---|---|---|
| Docling | CPU | 26,161 | 317 | |
| Docling | GPU | 26,160 | 96 | |
| Docling | SmolVLM (256M) | GPU | 28,285 | 257 |
| Docling | Granite Docling (258M) | GPU | 27,576 | 479 |
| Docling | SmolVLM (1.7B) | GPU | 28,361 | 204 |
| Docling | Qwen 3 (2B) | GPU | 28,427 | 247 |
| Docling | Qwen 3 (4B) | GPU | 28,883 | 312 |
Captioning Examples
SmolVLM (256M)
Government Pension Fund Global Annual report 2023
In this image we can see a boy running on the road. In the background, we can see the snow, trees, houses, fencing, road, and the sky.
Granite Docling (258M)
Government Pension Fund Global Annual report 2023
In this image I can see a child is standing on the road. On the left side there is a house. At the top there is snow.
SmolVLM (1.7B)
Government Pension Fund Global Annual report 2023
In the image, a young boy is captured mid-action. He is dressed in a blue winter outfit, including a puffy jacket and a matching beanie. The jacket has multiple pockets and buttons, suggesting it is a well-worn and functional piece of gear. The boy’s expression is joyful, as indicated by his open mouth and the playful nature of his movements.
He is standing on a snowy street, with a sidewalk to his left and a street to his right. The sidewalk is covered in snow, with tire tracks visible, suggesting that it has been recently cleared. The street is lined with houses, which are partially obscured by the snow. The houses are mostly white, with some having a touch of color, such as a red roof or a green window.
The background of the image is a blur of snow-covered trees and houses, indicating that the scene is set in a residential area. The sky is overcast, with a hint of sunlight peeking
Qwen 3 (2B)
Government Pension Fund Global Annual report 2023
A young child in a dark blue snowsuit and beanie is joyfully jumping over a puddle on a snow-covered street. The child’s face is lit with a smile as they leap, capturing a moment of pure, unbridled happiness in a winter scene. The background features snow-laden houses and trees, creating a peaceful, wintry atmosphere.
Qwen 3 (4B)
Government Pension Fund Global Annual report 2023
A joyful child in a dark blue snowsuit and knit hat leaps across a snowy street, kicking up snow as they run toward a puddle. The background features snow-covered houses, trees, and a wooden fence under a soft, overcast sky. The scene captures a moment of pure winter play and delight.
SmolVLM (256M)
In this image we can see a person holding a baby.
Granite Docling (258M)
In this image I can see a person hand and I can see a baby hand. Background is blurred.
SmolVLM (1.7B)
Image consists of a baby sleeping on a bed. Baby is being held by a person. Background is blurred.
Qwen 3 (2B)
A close-up captures a baby’s tiny hand gently held by an adult’s larger hand, both wrapped in soft, textured fabric. The baby is nestled in a white knitted blanket, and the scene is softly lit, creating a warm and tender atmosphere. The image is framed with a dark blue diagonal overlay on the upper right.
Qwen 3 (4B)
A close-up captures a tender moment as an adult’s hand gently holds a baby’s tiny fingers, emphasizing connection and care. The baby, wrapped in a soft, textured white blanket, sleeps peacefully with eyes closed. A bold diagonal navy blue stripe cuts across the upper right corner, adding a modern graphic contrast to the warm, intimate scene.
SmolVLM (256M)
Market value
15,765 billion kroner
The image is a line graph titled “2023.” The graph is segmented into three distinct years, each represented by a different color: blue, gray, and dark blue. The years are labeled with numbers, starting from 2013 and ending in 2023.
Granite Docling (258M)
Market value
15,765 billion kroner
In this image we can see a chart, in this chart we can see some numbers and some text.
SmolVLM (1.7B)
Market value
15,765 billion kroner
Blue bar graph with years 2013 to 2023.
Qwen 3 (2B)
Market value
15,765 billion kroner
This bar chart illustrates a steady and significant increase in a value from 2013 to 2023. The data shows a progression from 5,038 in 2013 to 15,765 in 2023, with the most recent year’s value being the highest. The growth is consistent, with each year’s bar being taller than the previous one, indicating a positive trend.
Qwen 3 (4B)
Market value
15,765 billion kroner
The bar chart displays annual values from 2013 to 2023, showing a generally upward trend with a sharp increase in 2023. Most years show steady growth, with values rising from 5,038 in 2013 to 15,765 in 2023. The bar for 2023 is highlighted in bright blue, indicating it is the highest value in the period, reaching 15,765.