OpenAI GPT-5.5 "Spud" Pretraining Complete: What We Know About the Incoming Frontier Model

OpenAI confirmed that GPT-5.5 (internally codenamed "Spud") completed pretraining on March 24, 2026. CEO Sam Altman described it as "a very strong model" that could "really accelerate the economy," while CTO Greg Brockman called it "two years of research" with a "big model feel." Based on the typical 3-6 week safety evaluation timeline, release is expected between mid-April and early June. Prediction markets assign 78% probability of release by April 30. GPT-5.5 is being designed as the backbone of OpenAI's "unified super-app"—collapsing ChatGPT, Codex (coding agent), deep research, memory, and browser capabilities into a single platform called Atlas. For context on the competitive pressure driving this: Anthropic hit $30 billion ARR this week, outpacing OpenAI's $25 billion. Meanwhile, OpenAI's Sora discontinuation freed GPU resources specifically for Spud training and inference at launch scale. GPT-4o is being fully retired from all plans on April 3—the same week this article was written—so all users are already on GPT-5.x infrastructure.