{"benchmark_id":"wmt23","benchmark_name":"WMT23","benchmark_description":"The Eighth Conference on Machine Translation (WMT23) benchmark evaluating machine translation systems across 8 language pairs (14 translation directions) including general, biomedical, literary, and low-resource language translation tasks. Features specialized shared tasks for quality estimation, metrics evaluation, sign language translation, and discourse-level literary translation with professional human assessment.","max_score":1.0,"categories":["language","healthcare"],"modality":"text","total_models":4,"entries":[{"rank":1,"model_id":"gemini-1.5-pro","model_name":"Gemini 1.5 Pro","organization_name":"Google","organization_id":"google","benchmark_score":0.751,"normalized_score":0.751,"verified":false,"self_reported":true,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"speed_rps":null,"context_window":null,"release_date":"2024-05-01","announcement_date":"2024-05-01","multimodal":true,"param_count":null,"is_new":false},{"rank":2,"model_id":"gemini-1.5-flash","model_name":"Gemini 1.5 Flash","organization_name":"Google","organization_id":"google","benchmark_score":0.741,"normalized_score":0.741,"verified":false,"self_reported":true,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"speed_rps":null,"context_window":null,"release_date":"2024-05-01","announcement_date":"2024-05-01","multimodal":true,"param_count":null,"is_new":false},{"rank":3,"model_id":"gemini-1.5-flash-8b","model_name":"Gemini 1.5 Flash 8B","organization_name":"Google","organization_id":"google","benchmark_score":0.726,"normalized_score":0.726,"verified":false,"self_reported":true,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"speed_rps":null,"context_window":null,"release_date":"2024-03-15","announcement_date":"2024-03-15","multimodal":true,"param_count":8000000000,"is_new":false},{"rank":4,"model_id":"gemini-1.0-pro","model_name":"Gemini 1.0 Pro","organization_name":"Google","organization_id":"google","benchmark_score":0.717,"normalized_score":0.717,"verified":false,"self_reported":false,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"speed_rps":null,"context_window":null,"release_date":"2024-02-15","announcement_date":"2024-02-15","multimodal":false,"param_count":null,"is_new":false}]}