Atlas

Benchmarks

← All benchmarks

OmniDocBench v1.5

Multimodal · 2025-09-25

OmniDocBench v1.5 evaluates end-to-end parsing of 1,355 diverse PDF pages. Its Overall score is the mean of text accuracy derived from normalized edit distance, table TEDS, and formula CDM.

Top models (higher is better)

ModelScore
Kimi K391.1
Gemini 3 Flash Preview90.4
Gemini 3 Pro Preview90.2
Claude Fable 589.8
GPT-5.589.4
Kimi K2.589.3
Qwen3-VL-235B-A22B-Instruct89.2
Gemini 2.5 Pro88.0
Opus 4.887.9
Qwen2.5-VL-72B-Instruct87.0
GPT-5.6 Sol85.8
GPT-5.2 (2025-12-11)85.8
InternVL3-78B80.3
GPT-4o (2024-08-06)75.0
Loading Atlas data…