The rumor mill is spinning — and this time there’s substance behind it: OpenAI’s next major model, internally codenamed ‘Spud,’ finished pre-training on March 24 at Stargate, the new data center in Abilene, Texas. Sam Altman himself said the launch was ‘a few weeks’ away.
What We Know
Facts first: pre-training is complete. Greg Brockman called Spud the result of ‘two years of research’ and ‘not an incremental improvement.’ Whether the model ships as GPT-5.5 or GPT-6 depends on the performance jump over GPT-5.4, according to OpenAI.
On Polymarket, the probability of a release by end of April sits at 78 percent. By end of June, it’s over 95 percent. The most likely window is May — based on OpenAI’s typical three-to-six-week cadence between pre-training completion and release.
What We Don’t Know
Here’s the caveat: there’s no architecture paper, no parameter count, no benchmark results, no pricing, and no confirmed name. An alleged leak named April 14 as launch day alongside a ‘super app’ merging ChatGPT, Codex, and the Atlas browser — but the source is unverified and the details are suspiciously precise.
Why This Still Matters
The timing is significant. Anthropic just unveiled Mythos Preview — a model that scores 93.9 percent on SWE-bench Verified and that Anthropic itself says is too powerful for public release. If OpenAI follows up now, we’re looking at the most intense model competition the AI industry has ever seen.
For us as users, this means: it’s about to get interesting. Whether GPT-6 drops in April or May, the next generation of AI models is imminent. And if Brockman is to be believed, it won’t be a small step.
Sources: