OpenAI Rumored to Launch "Astra" AI Model Codenamed "mewfour" Next Week
Leaks suggest OpenAI is preparing to launch "Astra" (codename "mewfour"), which is reportedly their largest pre-trained model since GPT-4.5. The model has reached the release candidate stage and is rumored to launch as early as next week. If the rumors are true, Astra represents a major leap in OpenAI's capabilities, particularly in handling long-horizon tasks and orchestrating multiple AI agents. This could accelerate the transition from simple chatbots to highly collaborative, autonomous AI agent workflows. An internal version of Astra reportedly solved 10 major open math problems, consuming approximately $2,000 worth of tokens based on Sol API rates. The model is specifically designed to coordinate and organize AI agents for complex, multi-step reasoning tasks.
## BACKGROUND
AI agent coordination, or orchestration, refers to the system design where multiple autonomous AI agents work together, sharing data and dividing tasks to achieve a collective goal. GPT-4.5 was OpenAI's previous benchmark for large-scale pre-training, and Astra represents the next step in scaling up model size and reasoning capabilities.