OpenAI’s GPT-5.4 Pro has been publicly credited with solving Erdős Problem #1196, a long-running problem about primitive sets, after being prompted by Liam Price. The strongest public evidence comes from the Erdős Problems website, which now marks the problem as solved and says it was solved by “GPT-5.4 Pro (prompted by Price)”.

The site says the model proved a stronger tail bound for primitive sets and lists the problem as “PROVED (LEAN),” indicating the proof was formalized and mechanically verified. In the related discussion thread, Price linked to a shared ChatGPT conversation that he identified as the chat that solved the problem. That shared page shows a single initial problem prompt and a visible reasoning duration of 80 minutes and 17 seconds.

What remains less clear is the viral social-media framing around the result. Accessible primary materials confirm the attribution to GPT-5.4 Pro and Price, but they do not independently establish every circulating claim about Price’s age, training background, or the broader narrative that the model broke through a decades-long mathematical “mental block” in one unprecedented leap. Still, the public record is strong enough to show that a frontier model is now being taken seriously in high-level mathematical problem solving, especially when prominent mathematicians discussing the proof describe the route as unusual and partly opaque.