{"benchmark_id":"exploitgym","benchmark_name":"ExploitGym","benchmark_description":"ExploitGym is a large-scale, realistic benchmark built from real-world vulnerabilities across userspace programs, Google's V8 engine, and the Linux kernel. Given a vulnerability and a proof-of-vulnerability input, agents must craft a working end-to-end exploit that achieves unauthorized code execution.","max_score":1.0,"categories":["safety","agents","code"],"modality":"text","total_models":8,"entries":[{"rank":1,"model_id":"gpt-6-astra","model_name":"GPT-6 Astra","organization_name":"OpenAI","organization_id":"openai","benchmark_score":0.424,"normalized_score":0.424,"verified":false,"self_reported":true,"provider_id":"openai","input_cost_per_million":10.0,"output_cost_per_million":50.0,"speed_rps":null,"context_window":1050000,"release_date":"2026-09-04","announcement_date":"2026-09-03","multimodal":true,"param_count":null,"is_new":false},{"rank":2,"model_id":"gpt-5.6-sol","model_name":"GPT-5.6 Sol","organization_name":"OpenAI","organization_id":"openai","benchmark_score":0.337,"normalized_score":0.337,"verified":false,"self_reported":true,"provider_id":"openai","input_cost_per_million":5.0,"output_cost_per_million":30.0,"speed_rps":null,"context_window":1050000,"release_date":"2026-07-09","announcement_date":"2026-07-09","multimodal":true,"param_count":null,"is_new":false},{"rank":3,"model_id":"gpt-5.6-terra","model_name":"GPT-5.6 Terra","organization_name":"OpenAI","organization_id":"openai","benchmark_score":0.232,"normalized_score":0.232,"verified":false,"self_reported":true,"provider_id":"openai","input_cost_per_million":2.0,"output_cost_per_million":12.0,"speed_rps":null,"context_window":1050000,"release_date":"2026-07-09","announcement_date":"2026-07-09","multimodal":true,"param_count":null,"is_new":false},{"rank":4,"model_id":"mimo-v2.6-pro","model_name":"MiMo-V2.6-Pro","organization_name":"Xiaomi","organization_id":"xiaomi","benchmark_score":0.178,"normalized_score":0.178,"verified":false,"self_reported":true,"provider_id":"xiaomi","input_cost_per_million":0.435,"output_cost_per_million":0.87,"speed_rps":null,"context_window":1048576,"release_date":"2026-09-22","announcement_date":"2026-09-22","multimodal":true,"param_count":1020000000000,"is_new":false},{"rank":5,"model_id":"deepseek-v4.1-flash","model_name":"DeepSeek-V4.1-Flash","organization_name":"DeepSeek","organization_id":"deepseek","benchmark_score":0.153,"normalized_score":0.153,"verified":false,"self_reported":true,"provider_id":"fireworks","input_cost_per_million":0.22,"output_cost_per_million":0.66,"speed_rps":null,"context_window":1040000,"release_date":"2026-09-10","announcement_date":"2026-09-10","multimodal":true,"param_count":763205315794,"is_new":false},{"rank":6,"model_id":"glm-5.3","model_name":"GLM-5.3","organization_name":"Zhipu AI","organization_id":"zai-org","benchmark_score":0.15,"normalized_score":0.15,"verified":false,"self_reported":true,"provider_id":"deepinfra","input_cost_per_million":1.2,"output_cost_per_million":4.0,"speed_rps":null,"context_window":1048576,"release_date":"2026-08-14","announcement_date":"2026-08-14","multimodal":false,"param_count":753000000000,"is_new":false},{"rank":7,"model_id":"gpt-5.6-luna","model_name":"GPT-5.6 Luna","organization_name":"OpenAI","organization_id":"openai","benchmark_score":0.124,"normalized_score":0.124,"verified":false,"self_reported":true,"provider_id":"openai","input_cost_per_million":0.2,"output_cost_per_million":1.2,"speed_rps":null,"context_window":1050000,"release_date":"2026-07-09","announcement_date":"2026-07-09","multimodal":true,"param_count":null,"is_new":false},{"rank":8,"model_id":"mimo-v2.6-flash","model_name":"MiMo-V2.6-Flash","organization_name":"Xiaomi","organization_id":"xiaomi","benchmark_score":0.06,"normalized_score":0.06,"verified":false,"self_reported":true,"provider_id":"xiaomi","input_cost_per_million":0.14,"output_cost_per_million":0.28,"speed_rps":null,"context_window":1048576,"release_date":"2026-09-22","announcement_date":"2026-09-22","multimodal":true,"param_count":309000000000,"is_new":false}]}