Sam Altman confirmed one thing on March 24: pre-training for their next large model is complete, just weeks away from release. The internal codename for this model is Spud — the English word for potato.
OpenAI has always been casual with development-stage model names (GPT-5 was Orion, and before that there was Strawberry), but what makes this interesting is not the name itself, but a follow-up comment from Greg Brockman.
What Greg Said
In an internal presentation, Greg Brockman described Spud with these words:
Two years of research are packed into this model. It's not a small iteration — it has a big model feel.
This statement is worth considering in the context of 2026.
Throughout the first half of 2026, OpenAI has been releasing models on a monthly cadence: GPT-5.2, GPT-5.3 Codex, GPT-5.4, along with their respective mini and nano variants. These updates are fast and steady, but each is a single-dimensional enhancement on the 5.x foundation — slightly better reasoning, stronger coding, more reliable tool calling.
What does two years of research mean? It means Spud is not just another decimal-point version. It may represent a qualitative change in some fundamental capabilities.
Sam Altman's own description: a truly powerful model that can accelerate the economy. That phrasing carries more weight than his usual remarks.
What This Model Might Offer
Based on information leaked so far:
- Native multimodality: Text, images, audio, and video unified in a single model, rather than being separate or bolted on as they are now
- Stronger agent capabilities: Autonomous execution of multi-step tasks. Since OpenAI shelved Sora, it has been concentrating resources on the enterprise agent direction
- Fewer hallucinations: GPT-5.4 already reduced factual errors by 33% compared to the previous generation; Spud aims to push this further
- Less need for prompt engineering: Greg mentioned better context understanding, meaning users won't have to explain everything — in short, you won't need to write complex system prompts; the model can figure out what you're doing on its own
As for whether the official product name will be GPT-5.5 or GPT-6, OpenAI itself hasn't decided yet, saying it depends on final performance evaluations. This detail is quite telling — if they themselves think it could be called 6, it suggests the capability leap is real.
Release Cadence: OpenAI's Monthly Model Releases in 2026
| Model | Release Date |
|---|---|
| GPT-5.2 | January 2026 |
| GPT-5.3 Codex | February 2026 |
| GPT-5.4 | March 5, 2026 |
| Spud (GPT-5.5/6) | April–May 2026 (estimated) |
This pace is unusually fast by historical standards. OpenAI used to release a new flagship every six months; in 2026, it has become monthly.
The likely reason: the competitive landscape has changed. DeepSeek, Anthropic, and Google are all catching up quickly, and OpenAI can no longer release at its own leisurely pace — first-mover advantage has become one of the most important variables in the market.
What Prediction Markets Are Betting On
On Polymarket, the market for whether GPT-5.5 will be released before June 30 currently shows odds exceeding 90% probability of a release by that date.
More aggressive bets suggest a significant chance of release before April 30. Combined with Sam Altman's mention of "weeks" (he said this on March 24), the math points to April or early May.
It's worth noting that OpenAI has never committed to anything on prediction markets, but each release cadence has largely matched market expectations — indicating that information leaks quite quickly, and the duration of information advantage is shrinking.
Competitors on the Same Stage
The models that Spud will face head-on in Q2 2026 are not at the level of a few months ago:
| Model | Company | Status |
|---|---|---|
| DeepSeek V4 | DeepSeek | Expected late April |
| Claude Mythos | Anthropic | Limited preview available, SWE-bench Pro 77.8% |
| Gemini 3.2 | Expected Q2 | |
| Grok 5 | xAI | Expected Q2 |
No one is naive; everyone is waiting for someone else to go first. Whoever releases first takes the first wave of benchmark comparisons.
OpenAI has historically favored striking first — Sam Altman's consistent logic is to release and iterate quickly rather than waiting for perfection. Spud will likely follow this approach.
The question is: when competitors are all releasing within the same window, how long does the first-mover card remain effective?
The Name "Spud"
As a side note: whoever is on OpenAI's naming committee, their style has remained consistent — Strawberry, Spud... They are using food names to chronicle the evolution of an era.
It's a bit absurd, but Spud is easier to remember than any version number.
Sources: GPT-5.5 Spud: Everything About OpenAI Next Frontier Model (pasqualepillitteri.it); GPT-5.5 Is Coming in 2026: Release Date, Features (Panstag); CocoLoop; AI Models in April 2026: Every Major Release, Leak, and What Comes Next (renovateqr.com)