{"benchmark_id":"mrcr-v2-(8-needle)","name":"MRCR v2 (8-needle)","parent_benchmark":null,"categories":["reasoning","long_context","general"],"modality":"text","multilingual":false,"max_score":1.0,"language":"en","description":"MRCR v2 (8-needle) is a variant of the Multi-Round Coreference Resolution benchmark that includes 8 needle items to retrieve from long contexts. This tests models' ability to simultaneously track and reason about multiple pieces of information across extended conversations.","paper_link":"https://arxiv.org/abs/2409.12640","implementation_link":null,"verified":false,"created_at":"2026-05-07T16:53:25.190272+00:00","updated_at":"2026-07-05T18:27:56.130932+00:00","statistics":{"total_models":10,"average_score":0.39449999999999996,"min_score":0.164,"max_score":0.76,"score_stddev":0.2204506747551479,"verified_count":0,"self_reported_count":10},"child_benchmarks":[],"linked_dataset":{"dataset_id":"eb062b1f-039b-4489-b61a-3d12c67a4851","slug":"mrcr-v2","organization_id":"8c4c8154-2985-4ddc-bee3-f156fbb9745b","display_name":"mrcr-v2","leaderboards":[{"id":"ac3b97e6-6c6b-4859-b608-b1cf8a1d7989","name":"mean_similarity leaderboard","metric_name":"mean_similarity","metric_source":"column_metric","subset_name":"8needle_upto_128k","sort_direction":"desc"},{"id":"689dcf18-687f-482e-bce5-0d102ae7ce2b","name":"mean_similarity leaderboard","metric_name":"mean_similarity","metric_source":"column_metric","subset_name":"8needle_4096_8192","sort_direction":"desc"},{"id":"9b0ecc97-0f41-46a0-81cd-798736af11c5","name":"mean_similarity leaderboard","metric_name":"mean_similarity","metric_source":"column_metric","subset_name":"8needle_32768_65536","sort_direction":"desc"}]},"models":[{"rank":1,"model_id":"claude-opus-4-6","model_name":"Claude Opus 4.6","organization_id":"anthropic","organization_name":"Anthropic","organization_country":"US","score":0.76,"normalized_score":0.76,"verified":false,"self_reported":true,"self_reported_source":"https://www.anthropic.com/news/claude-opus-4-6","analysis_method":"MRCR v2 1M 8-needles variant. Mean Match Ratio with max output tokens, average of 5 trials. Anthropic also reports 78.3 with 64k output tokens and 93.0 for the 256K 8-needle subset.","verification_date":null,"provider_id":"anthropic","input_cost_per_million":5.0,"output_cost_per_million":25.0,"context_window":1000000,"announcement_date":"2026-02-05","param_count":null,"is_open_source":false,"is_new":false,"best_latency":0.5,"latency_provider":"Anthropic","best_throughput":42.0,"throughput_provider":"Anthropic","context_provider":"Anthropic"},{"rank":2,"model_id":"gpt-5.5","model_name":"GPT-5.5","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.74,"normalized_score":0.74,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/introducing-gpt-5-5/","analysis_method":"OpenAI MRCR v2 8-needle, 512K-1M context range. Reasoning effort xhigh.","verification_date":null,"provider_id":"openai","input_cost_per_million":5.0,"output_cost_per_million":30.0,"context_window":1050000,"announcement_date":"2026-04-23","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":3,"model_id":"gemini-3.1-flash-lite-preview","model_name":"Gemini 3.1 Flash-Lite","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.601,"normalized_score":0.601,"verified":false,"self_reported":true,"self_reported_source":"https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/","analysis_method":"128k (average)","verification_date":null,"provider_id":"google","input_cost_per_million":0.25,"output_cost_per_million":1.5,"context_window":1000000,"announcement_date":"2026-03-03","param_count":null,"is_open_source":false,"is_new":false,"best_latency":null,"latency_provider":"Google","best_throughput":null,"throughput_provider":"Google","context_provider":"Google"},{"rank":4,"model_id":"gpt-5.4-mini","model_name":"GPT-5.4 mini","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.336,"normalized_score":0.336,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/introducing-gpt-5-4-mini-and-nano/","analysis_method":"OpenAI MRCR v2 8-needle, 128K-256K context range.","verification_date":null,"provider_id":"openai","input_cost_per_million":0.75,"output_cost_per_million":4.5,"context_window":400000,"announcement_date":"2026-03-17","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":5,"model_id":"gpt-5.4-nano","model_name":"GPT-5.4 nano","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.331,"normalized_score":0.331,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/introducing-gpt-5-4-mini-and-nano/","analysis_method":"OpenAI MRCR v2 8-needle, 128K-256K context range.","verification_date":null,"provider_id":"openai","input_cost_per_million":0.2,"output_cost_per_million":1.25,"context_window":400000,"announcement_date":"2026-03-17","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":6,"model_id":"gemini-3.5-flash","model_name":"Gemini 3.5 Flash","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.266,"normalized_score":0.266,"verified":false,"self_reported":true,"self_reported_source":"https://deepmind.google/models/evals-methodology/gemini-3-5-flash/","analysis_method":"1M (pointwise)","verification_date":null,"provider_id":"google","input_cost_per_million":1.5,"output_cost_per_million":9.0,"context_window":1048576,"announcement_date":"2026-05-19","param_count":null,"is_open_source":false,"is_new":false,"best_latency":null,"latency_provider":"Google","best_throughput":null,"throughput_provider":"Google","context_provider":"Google"},{"rank":7,"model_id":"gemini-3-pro-preview","model_name":"Gemini 3 Pro","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.263,"normalized_score":0.263,"verified":false,"self_reported":true,"self_reported_source":"https://blog.google/products/gemini/gemini-3","analysis_method":"1M (pointwise)","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-11-18","param_count":null,"is_open_source":false,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null},{"rank":7,"model_id":"gemini-3.1-pro-preview","model_name":"Gemini 3.1 Pro","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.263,"normalized_score":0.263,"verified":false,"self_reported":true,"self_reported_source":"https://deepmind.google/models/evals-methodology/gemini-3-1-pro","analysis_method":"1M (pointwise)","verification_date":null,"provider_id":"google","input_cost_per_million":2.5,"output_cost_per_million":15.0,"context_window":1048576,"announcement_date":"2026-02-19","param_count":null,"is_open_source":false,"is_new":false,"best_latency":0.6,"latency_provider":"Google","best_throughput":90.0,"throughput_provider":"Google","context_provider":"Google"},{"rank":9,"model_id":"gemini-3-flash-preview","model_name":"Gemini 3 Flash","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.221,"normalized_score":0.221,"verified":false,"self_reported":true,"self_reported_source":"https://blog.google/products/gemini/gemini-3-flash/","analysis_method":"1M (pointwise)","verification_date":null,"provider_id":"google","input_cost_per_million":0.5,"output_cost_per_million":3.0,"context_window":1000000,"announcement_date":"2025-12-17","param_count":null,"is_open_source":false,"is_new":false,"best_latency":null,"latency_provider":"Google","best_throughput":null,"throughput_provider":"Google","context_provider":"Google"},{"rank":10,"model_id":"gemini-2.5-pro-preview-06-05","model_name":"Gemini 2.5 Pro Preview 06-05","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.164,"normalized_score":0.164,"verified":false,"self_reported":true,"self_reported_source":"https://blog.google/products/gemini/gemini-2-5-pro-latest-preview/","analysis_method":"1M pointwise","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-06-05","param_count":null,"is_open_source":false,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null}]}