AI Foundations · The 2027 model lineup
You can describe what OpenAI's public signals say about the next mainline model, distinguish the "bigger model" frame from the "different relationship" frame, and read prediction markets honestly.
Refreshed: May 2026.
The question "when is GPT-6?" is half a year old and getting louder. The first thing this chapter does is unbend the question.
The Spud surprise. Through 2025 and into early 2026, the long-rumored mainline successor to GPT-5 was codenamed "Spud." Most industry coverage expected Spud to ship as GPT-6. Instead, OpenAI shipped it as GPT-5.5 on April 23, 2026. The version number was a deliberate signal — what came out of the Spud training run was a strong incremental upgrade (FrontierMath 0.476, GPQA 92%), not the discontinuous step that the GPT-N convention has historically marked. OpenAI did not call it GPT-6. That tells you something about what they're saving the number for.
The forecasting layer. Prediction markets and forecasting tournaments are useful for timing uncertainty, but they are not model-roadmap evidence. The durable lesson from prediction-market research is that prices can aggregate dispersed expectations; the durable lesson from IARPA's ACE program is that eliciting, weighting, and combining many forecasts can improve probabilistic judgment. Lesson 1.6.2.3 walks how to use those signals without mistaking them for an OpenAI announcement.
Chapter contains 4 lessons.