{"benchmark_id":"perceptionbench","benchmark_name":"PerceptionBench","benchmark_description":"PerceptionBench is Moonshot AI's internal benchmark for evaluating atomic visual perception capabilities.","max_score":1.0,"categories":["multimodal","reasoning","vision"],"modality":"multimodal","total_models":2,"entries":[{"rank":1,"model_id":"qwen3.8-max","model_name":"Qwen3.8 Max","organization_name":"Alibaba Cloud / Qwen Team","organization_id":"qwen","benchmark_score":0.635,"normalized_score":0.635,"verified":false,"self_reported":true,"provider_id":"deepinfra","input_cost_per_million":1.65,"output_cost_per_million":4.951,"speed_rps":null,"context_window":256000,"release_date":"2026-08-02","announcement_date":"2026-08-02","multimodal":true,"param_count":2400000000000,"is_new":false},{"rank":2,"model_id":"kimi-k3","model_name":"Kimi K3","organization_name":"Moonshot AI","organization_id":"moonshotai","benchmark_score":0.585,"normalized_score":0.585,"verified":false,"self_reported":true,"provider_id":"novita","input_cost_per_million":3.0,"output_cost_per_million":15.0,"speed_rps":null,"context_window":1048576,"release_date":"2026-07-16","announcement_date":"2026-07-16","multimodal":true,"param_count":2800000000000,"is_new":false}]}