{"benchmark_id":"arena-hard-v2","name":"Arena-Hard v2","parent_benchmark":null,"categories":["reasoning","general","creativity","writing"],"modality":"text","multilingual":false,"max_score":1.0,"language":"en","description":"Arena-Hard-Auto v2 is a challenging benchmark consisting of 500 carefully curated prompts sourced from Chatbot Arena and WildChat-1M, designed to evaluate large language models on real-world user queries. The benchmark covers diverse domains including open-ended software engineering problems, mathematics, creative writing, and technical problem-solving. It uses LLM-as-a-Judge for automatic evaluation, achieving 98.6% correlation with human preference rankings while providing 3x higher separation of model performances compared to MT-Bench. The benchmark emphasizes prompt specificity, complexity, and domain knowledge to better distinguish between model capabilities.","paper_link":"https://arxiv.org/abs/2406.11939","implementation_link":null,"verified":false,"created_at":"2026-05-07T16:53:21.763691+00:00","updated_at":"2026-07-05T18:27:43.343627+00:00","statistics":{"total_models":16,"average_score":0.6608625,"min_score":0.368,"max_score":0.862,"score_stddev":0.13713023918888204,"verified_count":0,"self_reported_count":16},"child_benchmarks":[],"linked_dataset":null,"models":[{"rank":1,"model_id":"mimo-v2-flash","model_name":"MiMo-V2-Flash","organization_id":"xiaomi","organization_name":"Xiaomi","organization_country":"CN","score":0.862,"normalized_score":0.862,"verified":false,"self_reported":true,"self_reported_source":"https://mimo.xiaomi.com/blog/mimo-v2-flash","analysis_method":"Creative Writing","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-12-16","param_count":309000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":2,"model_id":"qwen3-next-80b-a3b-instruct","model_name":"Qwen3-Next-80B-A3B-Instruct","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.827,"normalized_score":0.827,"verified":false,"self_reported":true,"self_reported_source":"https://qwenlm.github.io/blog/qwen3-next/","analysis_method":"GPT-4.1 evaluated win rates","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-09-10","param_count":80000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":3,"model_id":"qwen3-235b-a22b-thinking-2507","model_name":"Qwen3-235B-A22B-Thinking-2507","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.797,"normalized_score":0.797,"verified":false,"self_reported":true,"self_reported_source":"https://qwenlm.github.io/blog/qwen3-thinking/","analysis_method":"GPT-4 evaluated win rates","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-07-25","param_count":235000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":4,"model_id":"qwen3-235b-a22b-instruct-2507","model_name":"Qwen3-235B-A22B-Instruct-2507","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.792,"normalized_score":0.792,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507","analysis_method":"Win Rate","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-07-22","param_count":235000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":5,"model_id":"qwen3-vl-235b-a22b-instruct","model_name":"Qwen3 VL 235B A22B Instruct","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.774,"normalized_score":0.774,"verified":false,"self_reported":true,"self_reported_source":null,"analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-09-22","param_count":236000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":6,"model_id":"nemotron-3-super-120b-a12b","model_name":"Nemotron 3 Super (120B A12B)","organization_id":"nvidia","organization_name":"NVIDIA","organization_country":"US","score":0.7388,"normalized_score":0.7388,"verified":false,"self_reported":true,"self_reported_source":"https://build.nvidia.com/nvidia/nemotron-3-super-120b-a12b/modelcard","analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-03-11","param_count":120000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":7,"model_id":"sarvam-105b","model_name":"Sarvam-105B","organization_id":"sarvamai","organization_name":"Sarvam AI","organization_country":"IN","score":0.71,"normalized_score":0.71,"verified":false,"self_reported":true,"self_reported_source":"https://www.sarvam.ai/blogs/sarvam-30b-105b","analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-03-06","param_count":105000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":8,"model_id":"nemotron-3-nano-30b-a3b","model_name":"Nemotron 3 Nano (30B A3B)","organization_id":"nvidia","organization_name":"NVIDIA","organization_country":"US","score":0.677,"normalized_score":0.677,"verified":false,"self_reported":true,"self_reported_source":"https://build.nvidia.com/nvidia/nemotron-3-nano-30b-a3b/modelcard","analysis_method":"Average","verification_date":null,"provider_id":"deepinfra","input_cost_per_million":0.06,"output_cost_per_million":0.24,"context_window":262144,"announcement_date":"2025-12-15","param_count":32000000000,"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":"qwen3-vl-32b-instruct","model_name":"Qwen3 VL 32B Instruct","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.647,"normalized_score":0.647,"verified":false,"self_reported":true,"self_reported_source":null,"analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-09-22","param_count":33000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":10,"model_id":"qwen3-next-80b-a3b-thinking","model_name":"Qwen3-Next-80B-A3B-Thinking","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.623,"normalized_score":0.623,"verified":false,"self_reported":true,"self_reported_source":"https://qwenlm.github.io/blog/qwen3-next/","analysis_method":"GPT-4.1 evaluated win rates","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-09-10","param_count":80000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":11,"model_id":"qwen3-vl-32b-thinking","model_name":"Qwen3 VL 32B Thinking","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.605,"normalized_score":0.605,"verified":false,"self_reported":true,"self_reported_source":null,"analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-09-22","param_count":33000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":12,"model_id":"qwen3-vl-30b-a3b-instruct","model_name":"Qwen3 VL 30B A3B Instruct","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.585,"normalized_score":0.585,"verified":false,"self_reported":true,"self_reported_source":null,"analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-09-22","param_count":31000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":13,"model_id":"qwen3-vl-30b-a3b-thinking","model_name":"Qwen3 VL 30B A3B Thinking","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.567,"normalized_score":0.567,"verified":false,"self_reported":true,"self_reported_source":null,"analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-09-22","param_count":31000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":14,"model_id":"qwen3-vl-8b-thinking","model_name":"Qwen3 VL 8B Thinking","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.511,"normalized_score":0.511,"verified":false,"self_reported":true,"self_reported_source":null,"analysis_method":"winrate","verification_date":null,"provider_id":"deepinfra","input_cost_per_million":0.18,"output_cost_per_million":2.09,"context_window":262144,"announcement_date":"2025-09-22","param_count":9000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"DeepInfra","best_throughput":null,"throughput_provider":"DeepInfra","context_provider":"DeepInfra"},{"rank":15,"model_id":"sarvam-30b","model_name":"Sarvam-30B","organization_id":"sarvamai","organization_name":"Sarvam AI","organization_country":"IN","score":0.49,"normalized_score":0.49,"verified":false,"self_reported":true,"self_reported_source":"https://www.sarvam.ai/blogs/sarvam-30b-105b","analysis_method":null,"verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-03-06","param_count":30000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":16,"model_id":"qwen3-vl-4b-thinking","model_name":"Qwen3 VL 4B Thinking","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.368,"normalized_score":0.368,"verified":false,"self_reported":true,"self_reported_source":null,"analysis_method":"winrate","verification_date":null,"provider_id":"deepinfra","input_cost_per_million":0.1,"output_cost_per_million":1.0,"context_window":262144,"announcement_date":"2025-09-22","param_count":4000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"DeepInfra","best_throughput":null,"throughput_provider":"DeepInfra","context_provider":"DeepInfra"}]}