{"benchmark_id":"tau3-bench","name":"TAU3-Bench","parent_benchmark":null,"categories":["reasoning","agents","tool_calling"],"modality":"text","multilingual":false,"max_score":1.0,"language":"en","description":"TAU3-Bench is a benchmark for evaluating general-purpose agent capabilities, testing models on multi-turn interactions with simulated user models, retrieval, and complex decision-making scenarios.","paper_link":null,"implementation_link":null,"verified":false,"created_at":"2026-05-07T16:53:26.462208+00:00","updated_at":"2026-10-31T18:28:39.275499+00:00","statistics":{"total_models":9,"average_score":0.6178222222222223,"min_score":0.226,"max_score":0.862,"score_stddev":0.18410883858317195,"verified_count":0,"self_reported_count":9},"child_benchmarks":[],"linked_dataset":null,"models":[{"rank":1,"model_id":"gpt-live-1","model_name":"GPT-Live-1","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.862,"normalized_score":0.862,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/introducing-gpt-live-1-in-the-api/","analysis_method":"Tau3 (Voice) Intelligence, Pass@1, averaged across airline/retail/telecom domains. GPT-Live-1 backend: GPT-6 Astra (medium). Official OpenAI launch blog table.","verification_date":null,"provider_id":"openai","input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-09-10","param_count":null,"is_open_source":false,"is_new":false,"best_latency":null,"latency_provider":"OpenAI","best_throughput":null,"throughput_provider":"OpenAI","context_provider":"OpenAI"},{"rank":2,"model_id":"mimo-v2.5-pro","model_name":"MiMo-V2.5-Pro","organization_id":"xiaomi","organization_name":"Xiaomi","organization_country":"CN","score":0.729,"normalized_score":0.729,"verified":false,"self_reported":true,"self_reported_source":"https://mimo.xiaomi.com/mimo-v2-5-pro/","analysis_method":null,"verification_date":null,"provider_id":"xiaomi","input_cost_per_million":0.435,"output_cost_per_million":0.87,"context_window":1048576,"announcement_date":"2026-04-27","param_count":1023244718976,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"Xiaomi","best_throughput":null,"throughput_provider":"Xiaomi","context_provider":"DeepInfra"},{"rank":3,"model_id":"qwen3.6-plus","model_name":"Qwen3.6 Plus","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.707,"normalized_score":0.707,"verified":false,"self_reported":true,"self_reported_source":"https://qwen.ai/blog?id=qwen3.6","analysis_method":null,"verification_date":null,"provider_id":"novita","input_cost_per_million":0.5,"output_cost_per_million":3.0,"context_window":1000000,"announcement_date":"2026-04-02","param_count":null,"is_open_source":false,"is_new":false,"best_latency":null,"latency_provider":"Novita","best_throughput":null,"throughput_provider":"Novita","context_provider":"Novita"},{"rank":4,"model_id":"glm-5.1","model_name":"GLM-5.1","organization_id":"zai-org","organization_name":"Zhipu AI","organization_country":"CN","score":0.706,"normalized_score":0.706,"verified":false,"self_reported":true,"self_reported_source":"https://z.ai/blog/glm-5.1","analysis_method":null,"verification_date":null,"provider_id":"deepinfra","input_cost_per_million":1.05,"output_cost_per_million":3.5,"context_window":202752,"announcement_date":"2026-04-07","param_count":754000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"ZAI","best_throughput":null,"throughput_provider":"ZAI","context_provider":"DeepInfra"},{"rank":5,"model_id":"qwen3.6-35b-a3b","model_name":"Qwen3.6-35B-A3B","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.672,"normalized_score":0.672,"verified":false,"self_reported":true,"self_reported_source":"https://qwen.ai/blog?id=qwen3.6-35b-a3b","analysis_method":"Official user model (gpt-5.2 low) + BM25 retrieval","verification_date":null,"provider_id":"deepinfra","input_cost_per_million":0.1,"output_cost_per_million":0.95,"context_window":262144,"announcement_date":"2026-04-16","param_count":35000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"DeepInfra","best_throughput":null,"throughput_provider":"DeepInfra","context_provider":"DeepInfra"},{"rank":6,"model_id":"granite-4.2-30b","model_name":"IBM Granite 4.2 30B","organization_id":"ibm","organization_name":"IBM","organization_country":"US","score":0.62,"normalized_score":0.62,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/ibm-granite/granite-4.2-30b","analysis_method":"Score","verification_date":null,"provider_id":"deepinfra","input_cost_per_million":0.16,"output_cost_per_million":0.65,"context_window":131072,"announcement_date":"2026-08-25","param_count":30000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"DeepInfra","best_throughput":null,"throughput_provider":"DeepInfra","context_provider":"DeepInfra"},{"rank":7,"model_id":"granite-4.2-8b","model_name":"IBM Granite 4.2 8B","organization_id":"ibm","organization_name":"IBM","organization_country":"US","score":0.5806,"normalized_score":0.5806,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/ibm-granite/granite-4.2-8b","analysis_method":"Score","verification_date":null,"provider_id":"deepinfra","input_cost_per_million":0.06,"output_cost_per_million":0.25,"context_window":131072,"announcement_date":"2026-08-25","param_count":8000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"DeepInfra","best_throughput":null,"throughput_provider":"DeepInfra","context_provider":"DeepInfra"},{"rank":8,"model_id":"granite-4.2-3b","model_name":"IBM Granite 4.2 3B","organization_id":"ibm","organization_name":"IBM","organization_country":"US","score":0.4578,"normalized_score":0.4578,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/ibm-granite/granite-4.2-3b","analysis_method":"Score","verification_date":null,"provider_id":"deepinfra","input_cost_per_million":0.03,"output_cost_per_million":0.12,"context_window":131072,"announcement_date":"2026-08-25","param_count":3000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"DeepInfra","best_throughput":null,"throughput_provider":"DeepInfra","context_provider":"DeepInfra"},{"rank":9,"model_id":"nemotron-3-ultra-550b-a55b","model_name":"Nemotron 3 Ultra (550B A55B)","organization_id":"nvidia","organization_name":"NVIDIA","organization_country":"US","score":0.226,"normalized_score":0.226,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16","analysis_method":"Banking","verification_date":null,"provider_id":"deepinfra","input_cost_per_million":0.5,"output_cost_per_million":2.2,"context_window":262144,"announcement_date":"2026-06-04","param_count":550000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"DeepInfra","best_throughput":null,"throughput_provider":"DeepInfra","context_provider":"DeepInfra"}]}