Comparing parsers on some tags
Pass rate minus median of the reported systems.
table
math
form
multi column
multilingual
handwriting
rotated
tiny
degraded
list
heading
weights
Datalab accurate
closed
+9.0
+3.7
+10.9
+12.2
+25.8
+15.4
+27.4
+15.0
+19.5
+16.5
+15.9
Gemini 3.8 Flash*
closed
+11.0
+5.2
+9.2
+11.1
+25.6
+20.6
+28.3
+13.9
+19.4
+14.1
+16.8
Reducto
closed
+7.8
+3.9
+6.9
+8.5
+6.3
+8.8
+19.2
+12.7
+14.2
+5.6
+11.9
LlamaParse
closed
+9.2
+0.2
+1.8
+9.8
+21.9
+14.5
+26.8
+9.4
+13.3
+13.0
+12.6
chandra-ocr-2
open
+4.6
+3.4
+2.2
+8.5
+16.3
+9.5
+8.3
+13.1
+16.2
+13.9
+14.6
Mistral OCR
closed
+7.9
-0.4
+2.3
+4.5
+16.7
+6.7
+24.3
-8.3
+3.3
-2.9
+7.9
Claude Sonnet 5.5*
closed
+10.7
+6.4
+11.7
+8.4
+23.7
+17.1
+28.0
+15.1
+18.9
+15.4
+11.6
dots.mocr
open
+4.9
+1.1
-7.0
+2.6
+18.7
-3.6
+13.7
+0.0
-4.4
+4.9
+7.3
GPT-5.6 Sol
closed
+7.8
-0.5
+4.2
+5.9
+15.3
+9.4
+12.6
+9.6
+12.7
+9.4
+13.6
MinerU2.5-Pro
open
+8.2
+1.7
+0.0
+1.7
-6.7
-19.2
+19.7
+7.5
-4.1
+0.0
-5.3
Infinity-Parser2-Pro
open
+1.5
+3.3
+0.6
+6.9
+2.3
+14.1
-0.1
+7.5
+14.6
+3.6
+2.8
Extend
closed
+9.4
-51.8
+8.0
+2.7
+10.1
+15.2
+23.7
+10.2
+12.5
+2.8
+4.8
Surya OCR 2
open
-6.8
+1.6
-8.2
-1.3
+16.2
-7.6
-17.2
-2.5
-8.1
+6.0
+9.5
Azure
closed
-0.9
-4.6
+0.3
-0.8
-4.7
-18.5
+15.4
+8.5
-9.3
-0.2
-4.0
LightOnOCR-3-4B
open
-8.5
+4.3
+3.8
+2.6
+0.0
+9.9
-1.3
+8.1
+13.6
+2.9
+6.8
LightOnOCR-3-0.8B
open
-7.8
+4.0
+2.8
+0.0
-6.0
+0.0
+0.0
+4.9
+5.0
+0.0
+0.0
GLM-OCR
open
+0.9
+0.0
-13.0
-2.0
-20.7
-12.6
-1.0
-11.4
-5.2
-14.3
-16.6
Unlimited-OCR
open
-6.7
-0.9
-11.9
-7.6
-11.4
-23.5
-44.2
-17.7
-10.1
-15.2
-10.0
PaddleOCR-VL-1.6
open
+0.0
-26.0
-13.6
-9.4
-9.4
-15.0
+7.5
-18.3
-15.1
-18.7
-10.9
Infinity-Parser2-Flash
open
-14.2
+3.6
-6.1
-0.4
-10.4
+6.0
-27.1
-4.6
+8.7
-4.1
-5.6
DeepSeek-OCR-2
open
-12.6
-8.1
-21.9
-10.9
-7.3
-26.0
-49.2
-25.1
-21.1
-15.4
-5.3
Indra OCR
open
-20.8
-32.8
-23.4
-15.2
+9.5
-45.2
-12.5
-6.0
-33.6
-3.3
-3.3
HunyuanOCR-1.5
open
-19.4
-51.2
-24.0
-14.8
-9.2
-17.0
-19.1
-25.5
-12.1
-21.3
-8.9
olmOCR-2
open
-25.1
-5.5
-11.5
-12.1
-22.4
+0.8
-6.9
-22.1
+0.0
-11.7
-11.3
Nanonets-OCR2-3B
open
-19.7
-13.3
-7.6
-17.0
-17.4
-5.9
-35.4
-33.5
-5.9
-26.9
-7.0
median
83.7
89.8
81.7
80.5
60.0
60.8
64.7
79.7
65.4
78.9
69.8
A test counts toward every tag it has; multilingual is any language but English. The last row is the median itself.
* Used in creating and verifying at least some tests.