List your company

Desktop UI Navigation on OSWorld 50 easy tasks 1.0 (test)

100ASRClaude 3.5 Sonnet
79.284.69095.4Nov 4, 2024

Evaluation results

MethodLinks
Claude 3.5 SonnetObservation Modality=Screenshot, Action Step Limit=15, Decoding Temperature=02024.11100022
Claude 3.5 Sonnet v2Observation Modality=Screenshot, Action Step Limit=15, Decoding Temperature=02024.1196422
GPT-4oObservation Modality=Screenshot, Action Step Limit=15, Decoding Temperature=02024.1195.868
Claude 3.5 SonnetObservation Modality=SoM, Action Step Limit=15, Decoding Temperature=02024.1195.3644
Claude 3.5 Sonnet v2Observation Modality=SoM, Action Step Limit=15, Decoding Temperature=02024.1194.81058
GPT-4-TurboObservation Modality=Screenshot, Action Step Limit=15, Decoding Temperature=02024.1193.3218
GPT-4-TurboObservation Modality=SoM, Action Step Limit=15, Decoding Temperature=02024.1191.8852
GPT-4oObservation Modality=SoM, Action Step Limit=15, Decoding Temperature=02024.1191.226
Gemini 1.5Observation Modality=SoM, Action Step Limit=15, Decoding Temperature=02024.1188.7618
Gemini 1.5Observation Modality=Screenshot, Action Step Limit=15, Decoding Temperature=02024.118046