Overview
- OpenAI released multiple GPT-5.2 variants—instant, reasoning, professional, and the 'Thinking' model—positioned as its most capable to date.
- The company says the 'Thinking' version reduces factual errors by 38% compared with its predecessor.
- Applications chief Fidji Simo said Sam Altman's early-December 'red alert' increased resources that helped finalize the model, while disputing that the launch was accelerated.
- OpenAI cites 800 million weekly ChatGPT users, while Google reports 650 million monthly Gemini users as the rivalry intensifies.
- OpenAI has committed roughly $1.4 trillion over eight years for additional compute, is charging for Sora video generation, and targets an age-verified adult mode in Q1 2026 while not expecting profitability before 2029.