Skip to content

Benchmark

Mdpbench

Scores are percentages. Higher is better.

Models measured
10
Scores recorded
10
Sources
1
IQ weight
1.49

Score distribution

73.7%
Mean
78.9%
Median
9.9%
Std dev
63.9%
10th percentile
83.6%
90th percentile
51.8%
Min
51.8%83.6%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.49 for chat models, measured over 6 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Kimi K3· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim83.6%
2Qwen3.5 397B A17B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim83.6%
3Inkling 2 Small· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim81.4%
4Qwen 3.8 Flash Next· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim80.5%
5PaddleOCR-VL-1.6· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim78.9%
6Kimi K2.5· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim77.5%
7Qwen3 VL 8B Instruct· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim68.3%
8GLM OCR· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim67.3%
9LightOnOCR-2-1B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim63.9%
10DeepSeek-OCR· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim51.8%

Data credits: Hugging Face official benchmark leaderboards; licence varies by benchmark