OpenAI Fights Back: GPT-5.2 Unveiled to Rival Google’s Gemini 3 Pro

OpenAI unveiled GPT-5.2 on December 12, 2025, the model it says is its most capable yet for real‑world professional applications, developed to challenge Google’s Gemini 3 Pro, long-rumoured to be its flagship codenamed “Red Alert.” The release introduces three variants—Instant, Thinking, and Pro—initially available to paid subscribers, with GPT-5.2 Thinking positioned for deep reasoning and claiming a perfect 100% score on the AIME 2025 mathematics benchmark (GPT-5.1 scored 94%), and more than a 10% edge over GPT-5.1 on ARC-AGI-1 for abstract reasoning, alongside a reported 30% reduction in factual errors.

GPT-5.2 Instant remains focused on faster information retrieval, instructional guidance, and writing tasks. The timing follows Google’s Gemini 3 Pro securing top place on the LMArena rankings in November, and comes as OpenAI seeks to protect its market position after a USD 1.4 trillion infrastructure partnership and amid a broader scramble for productivity‑driven AI tools, according to OpenAI.

Altman’s memo and accompanying essay “Ten Years” frame the move as part of a longer mission toward AGI, even as the market watches how GPT-5.2 will perform against rivals.