A short-lived GitHub repository believed to be linked to OpenAI briefly published technical details for a forthcoming GPT-5 family of large language models before the page was removed. The posting listed four variants—GPT-5, GPT-5-mini, GPT-5-nano and GPT-5-chat—describing their intended uses from complex reasoning tasks to low-latency applications and enterprise-grade conversational systems. The leaked documentation said the flagship GPT-5 model can accept up to 400,000 input tokens and return as many as 128,000 output tokens, with the API supporting requests of up to 272,000 input tokens. The models allegedly integrate web search, image generation and multi-modal capabilities, while also offering fine-tuning options for developers. Pricing and access appear set to follow a tiered structure inside ChatGPT. Free users would receive 10 GPT-5 messages every five hours and one GPT-5-Thinking request daily before the service falls back to GPT-5-mini, while Plus subscribers would be allowed 80 messages every three hours and up to 200 GPT-5-Thinking prompts a week. Teams and Pro accounts are shown as having unlimited access.