{"benchmark_id":"mrcr-v2","name":"MRCR v2","parent_benchmark":null,"categories":["long_context","reasoning","general"],"modality":"text","multilingual":false,"max_score":1.0,"language":"en","description":"MRCR v2 (Multi-Round Coreference Resolution version 2) is an enhanced version of the synthetic long-context reasoning task. It extends the original MRCR framework with improved evaluation criteria and additional complexity for testing models' ability to maintain attention and reasoning across extended contexts.","paper_link":"https://arxiv.org/abs/2409.12640","implementation_link":null,"verified":false,"created_at":"2026-05-07T16:53:25.180175+00:00","updated_at":"2026-09-02T17:08:12.313816+00:00","statistics":{"total_models":3,"average_score":0.4676666666666667,"min_score":0.166,"max_score":0.917,"score_stddev":0.39667913145681527,"verified_count":0,"self_reported_count":3},"child_benchmarks":[],"linked_dataset":null,"models":[{"rank":1,"model_id":"qwen3.7-plus","model_name":"Qwen3.7-Plus","organization_id":"qwen","organization_name":"Alibaba Cloud / Qwen Team","organization_country":"CN","score":0.917,"normalized_score":0.917,"verified":false,"self_reported":true,"self_reported_source":"https://qwen.ai/blog?id=qwen3.7-plus","analysis_method":"128k","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-05-31","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":2,"model_id":"diffusiongemma-26b-a4b-it","model_name":"DiffusionGemma 26B-A4B","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.32,"normalized_score":0.32,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/google/diffusiongemma-26B-A4B-it","analysis_method":"8 needle 128k average","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2026-06-10","param_count":25200000000,"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":"gemini-2.5-flash-lite","model_name":"Gemini 2.5 Flash-Lite","organization_id":"google","organization_name":"Google","organization_country":"US","score":0.166,"normalized_score":0.166,"verified":false,"self_reported":true,"self_reported_source":"https://deepmind.google/models/gemini/flash-lite/","analysis_method":"Long context 128k average. 8 needle.","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-06-17","param_count":null,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":null,"best_throughput":null,"throughput_provider":null,"context_provider":null}]}