Polymarket traders are putting 75.5% odds on OpenAI releasing ChatGPT 5.5 this Wednesday, April 23. Leaked tests of the model, codenamed "Spud," show it building interactive 3D environments and outperforming Claude Opus 4.7 in both accuracy and versatility. If the prediction markets are right, OpenAI is about to drop its most significant model in over a year during the same week Apple announced a new CEO.

The timing is perfect, and probably intentional.

What the Leaks Actually Show

According to Universe of AI, which has been tracking Spud's development, the model demonstrates capabilities that go well beyond incremental improvement. The standout demos include a Minecraft-like 3D simulation with realistic physics and interactive elements, a detailed recreation of Monica's apartment from Friends using Three.js, professional-grade website designs generated from simple mockups, and SVG code generation with fewer lines and higher accuracy than any predecessor.

These are not chatbot tricks. This is a model building entire interactive environments from text prompts. If the leaked tests are representative of the final release, Spud represents the biggest single jump in consumer AI capability since GPT-4 launched in March 2023.

The Timeline Adds Up

Here is what we know about the development timeline. Pre-training for Spud was reportedly completed around March 24, 2026. OpenAI typically requires 3 to 6 weeks for safety evaluation before public release, which puts the earliest window at mid-April and the latest at early May. The model entered final safety review in the first week of April, according to Analytics Insight.

OpenAI has already started integrating Spud's capabilities into GPT 5.4 Pro for real-world testing, refining performance based on user feedback before the official release. Multiple outlets, including Digit and FindSkill, independently confirmed the April 21 to May 25 window, with April 23 emerging as the consensus target date. Thursday launches are OpenAI's preferred pattern.

Why This Launch Matters More Than Usual

OpenAI is under more competitive pressure right now than at any point in its history. Anthropic's Claude Opus 4.7 has been dominant in coding and reasoning benchmarks. Ramp data shows Anthropic overtaking OpenAI in enterprise AI spending for the first time in Q1 2026. Google's Gemini models are powering Apple's Siri. And OpenAI's stock exchange debut is looming.

Spud needs to do two things: prove OpenAI can still ship the best consumer AI model, and give the IPO roadshow something concrete to point to. The leaked 3D world generation demos accomplish both. They show a capability gap that neither Anthropic nor Google currently matches, and they target the creative and developer markets where engagement translates directly into subscription revenue.

The Caveat

Prediction markets are not announcements. Polymarket traders have been wrong before, and OpenAI has not confirmed any release date publicly. The company's most recent major model launch, GPT-5.4, arrived with little advance warning. Spud could follow the same pattern, or it could be delayed if safety evaluations surface issues.

But the convergence of signals is hard to ignore: pre-training complete, safety review in final stages, leaked demos showing dramatic capability improvements, integration testing already underway in production, and prediction market consensus forming around a specific date. If Spud does not ship this week, the next most likely window is late April to early May.

Either way, OpenAI needs this one to land. Anthropic is eating its enterprise lunch. The IPO clock is ticking. And the last time OpenAI's infrastructure made headlines, it was because ChatGPT went down for three hours and millions of people discovered they had no backup plan.

Sources: Polymarket (prediction data), Geeky Gadgets (leaked tests), Universe of AI (demo analysis), Digit, FindSkill, Analytics Insight, Reddit r/OpenAI.