{"benchmark_id":"collie","name":"COLLIE","parent_benchmark":null,"categories":["reasoning","language","writing"],"modality":"text","multilingual":false,"max_score":1.0,"language":"en","description":"COLLIE is a grammar-based framework for systematic construction of constrained text generation tasks. It allows specification of rich, compositional constraints across diverse generation levels and modeling challenges including language understanding, logical reasoning, and semantic planning. The COLLIE-v1 dataset contains 2,080 instances across 13 constraint structures.","paper_link":"https://arxiv.org/abs/2307.08689","implementation_link":null,"verified":false,"created_at":"2026-05-07T16:53:22.456530+00:00","updated_at":"2026-07-05T18:27:45.663813+00:00","statistics":{"total_models":10,"average_score":0.751,"min_score":0.425,"max_score":0.99,"score_stddev":0.21146420763597587,"verified_count":0,"self_reported_count":10},"child_benchmarks":[],"linked_dataset":null,"models":[{"rank":1,"model_id":"gpt-5-2025-08-07","model_name":"GPT-5","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.99,"normalized_score":0.99,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/gpt-5/","analysis_method":"GPT-5 with thinking mode enabled - Instruction-following in freeform writing.","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},{"rank":2,"model_id":"o3-mini","model_name":"o3-mini","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.987,"normalized_score":0.987,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/gpt-4-1/","analysis_method":"benchmark score","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-01-30","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":3,"model_id":"o3-2025-04-16","model_name":"o3","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.984,"normalized_score":0.984,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/gpt-5/","analysis_method":"OpenAI o3 with thinking mode enabled - Instruction-following in freeform writing.","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-04-16","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":4,"model_id":"mistral-medium-3-5","model_name":"Mistral Medium 3.5","organization_id":"mistral","organization_name":"Mistral AI","organization_country":"FR","score":0.958,"normalized_score":0.958,"verified":false,"self_reported":true,"self_reported_source":"https://huggingface.co/mistralai/Mistral-Medium-3.5-128B","analysis_method":"Collie. reasoning_effort=\"high\".","verification_date":null,"provider_id":"mistral","input_cost_per_million":1.5,"output_cost_per_million":7.5,"context_window":256000,"announcement_date":"2026-04-29","param_count":128000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"Mistral AI","best_throughput":null,"throughput_provider":"Mistral AI","context_provider":"Mistral AI"},{"rank":5,"model_id":"gpt-4.5","model_name":"GPT-4.5","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.723,"normalized_score":0.723,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/gpt-4-1/","analysis_method":"Accuracy","verification_date":null,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"context_window":null,"announcement_date":"2025-02-27","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":6,"model_id":"gpt-4.1-2025-04-14","model_name":"GPT-4.1","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.658,"normalized_score":0.658,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/introducing-gpt-5-for-developers/","analysis_method":"Standard benchmark","verification_date":null,"provider_id":"openai","input_cost_per_million":2.0,"output_cost_per_million":8.0,"context_window":1047576,"announcement_date":"2025-04-14","param_count":null,"is_open_source":false,"is_new":false,"best_latency":10.0,"latency_provider":"OpenAI","best_throughput":100.0,"throughput_provider":"OpenAI","context_provider":"OpenAI"},{"rank":7,"model_id":"mistral-small-latest","model_name":"Mistral Small 4","organization_id":"mistral","organization_name":"Mistral AI","organization_country":"FR","score":0.629,"normalized_score":0.629,"verified":false,"self_reported":true,"self_reported_source":"https://mistral.ai/news/mistral-small-4","analysis_method":"Collie. reasoning_effort=\"high\".","verification_date":null,"provider_id":"mistral","input_cost_per_million":0.15,"output_cost_per_million":0.6,"context_window":256000,"announcement_date":"2026-03-16","param_count":119000000000,"is_open_source":true,"is_new":false,"best_latency":null,"latency_provider":"Mistral AI","best_throughput":null,"throughput_provider":"Mistral AI","context_provider":"Mistral AI"},{"rank":8,"model_id":"gpt-4o-2024-08-06","model_name":"GPT-4o","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.61,"normalized_score":0.61,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/gpt-5/","analysis_method":"GPT-4o without thinking mode - Instruction-following in freeform writing.","verification_date":null,"provider_id":"openai","input_cost_per_million":2.5,"output_cost_per_million":10.0,"context_window":128000,"announcement_date":"2024-08-06","param_count":null,"is_open_source":false,"is_new":false,"best_latency":0.5,"latency_provider":"OpenAI","best_throughput":132.0,"throughput_provider":"OpenAI","context_provider":"OpenAI"},{"rank":9,"model_id":"gpt-4.1-mini-2025-04-14","model_name":"GPT-4.1 mini","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.546,"normalized_score":0.546,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/introducing-gpt-5-for-developers/","analysis_method":"Standard benchmark","verification_date":null,"provider_id":"openai","input_cost_per_million":0.4,"output_cost_per_million":1.6,"context_window":1047576,"announcement_date":"2025-04-14","param_count":null,"is_open_source":false,"is_new":false,"best_latency":5.0,"latency_provider":"OpenAI","best_throughput":150.0,"throughput_provider":"OpenAI","context_provider":"OpenAI"},{"rank":10,"model_id":"gpt-4.1-nano-2025-04-14","model_name":"GPT-4.1 nano","organization_id":"openai","organization_name":"OpenAI","organization_country":"US","score":0.425,"normalized_score":0.425,"verified":false,"self_reported":true,"self_reported_source":"https://openai.com/index/gpt-4-1/","analysis_method":"Standard benchmark","verification_date":null,"provider_id":"openai","input_cost_per_million":0.1,"output_cost_per_million":0.4,"context_window":1047576,"announcement_date":"2025-04-14","param_count":null,"is_open_source":false,"is_new":false,"best_latency":2.0,"latency_provider":"OpenAI","best_throughput":200.0,"throughput_provider":"OpenAI","context_provider":"OpenAI"}]}