ExtractBench Leaderboard

Open paper
Benchmark Model Rank Results
document-understanding-on-extractbenchClaude Code (Sonnet 5.5 Evidence)#2Value F1: 94.68
document-understanding-on-extractbenchCodex (GPT-5.6 Sol Evidence)#4Value F1: 93.77
document-understanding-on-extractbenchOpenAI GPT-6 Astra#5Value F1: 91.91
document-understanding-on-extractbenchCodex (GPT-5.6 Luna Evidence)#6Value F1: 91.16
document-understanding-on-extractbenchQwen3.8 27B#7Value F1: 89.75
document-understanding-on-extractbenchQwen3.6 35B#8Value F1: 88.11
document-understanding-on-extractbenchQwen3.5 35B#9Value F1: 88.0
document-understanding-on-extractbenchClaude Code (Opus 5.5 Evidence)#10Value F1: 86.06
document-understanding-on-extractbenchKimi K3#11Value F1: 83.17
document-understanding-on-extractbenchQwen3.5 4B#12Value F1: 82.43
document-understanding-on-extractbenchGLM-5.3 Flash#13Value F1: 80.75
document-understanding-on-extractbenchGoogle Gemini 3.8 Flash#14Value F1: 80.71
document-understanding-on-extractbenchGemma4 26B#15Value F1: 70.27
document-understanding-on-extractbenchGemma4 E4B#16Value F1: 69.64
document-understanding-on-extractbenchInternVL3.5 14B#17Value F1: 66.35
document-understanding-on-extractbenchQwen3.5 2B#18Value F1: 63.4
document-understanding-on-extractbenchGemma4 E2B#19Value F1: 51.85
document-understanding-on-extractbenchMiniCPM-V 4.5#20Value F1: 45.85
document-understanding-on-extractbenchGemma4 12B#21Value F1: 45.39
document-understanding-on-extractbenchGLM-4.6V Flash#22Value F1: 45.19
document-understanding-on-extractbenchClaude Code (Opus 4.8 Evidence)–Value F1: 86.84
document-understanding-on-extractbenchClaude Code (Opus 4.8)–Value F1: 87.09
document-understanding-on-extractbenchCodex (GPT-5.5 Evidence)–Value F1: 93.35
document-understanding-on-extractbenchCodex (GPT-5.5)–Value F1: 93.57
document-understanding-on-extractbenchCodex (GPT-6 Luna Evidence)–Value F1: 91.01
document-understanding-on-extractbenchGoogle Gemini 3.5 Flash–Value F1: 79.84
document-understanding-on-extractbenchLift Datalab 9B–Value F1: 78.43
document-understanding-on-extractbenchNuExtract 2.0 8B–Value F1: 49.42
document-understanding-on-extractbenchNuExtract3–Value F1: 81.02
document-understanding-on-extractbenchOpenAI GPT-5.4 Nano–Value F1: 74.9

Rank counts only results with a code link.