{"benchmark_id":"factscore","name":"FActScore","parent_benchmark":null,"categories":["reasoning"],"modality":"text","multilingual":false,"max_score":1.0,"language":"en","description":"A fine-grained atomic evaluation metric for factual precision in long-form text generation that breaks generated text into atomic facts and computes the percentage supported by reliable knowledge sources, with automated assessment using retrieval and language models","paper_link":"https://arxiv.org/abs/2305.14251","implementation_link":null,"verified":false,"created_at":"2026-05-07T16:53:23.072847+00:00","updated_at":"2026-07-05T18:27:47.670754+00:00","statistics":{"total_models":2,"average_score":0.49015000000000003,"min_score":0.01,"max_score":0.9703,"score_stddev":0.6790346419734417,"verified_count":0,"self_reported_count":2},"child_benchmarks":[],"linked_dataset":null,"models":[{"rank":1,"model_id":"grok-4.1-2025-11-17","model_name":"Grok-4.1","organization_id":"xai","organization_name":"xAI","organization_country":"US","score":0.9703,"normalized_score":0.9703,"verified":false,"self_reported":true,"self_reported_source":"https://x.ai/news/grok-4.1","analysis_method":"FActScore benchmark - 2.97% error rate (lower is better, converted to accuracy score). Evaluated on non-reasoning model with web search tools. Significant reduction from Grok 4 Fast's 9.89% error rate.","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-11-17","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":"gpt-5-2025-08-07","model_name":"GPT-5","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.01,"normalized_score":0.01,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/introducing-gpt-5-for-developers/","analysis_method":"Thinking mode enabled for factual accuracy assessment. Measured hallucination rate on open-source prompts.","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-08-07","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}]}