OpenAI Spud Drops Between April 14 and May 5 — 78% Polymarket, Greg Brockman Says 'Not Incremental': GPT-5.5 or GPT-6?
Spud, OpenAI's next flagship model, launches between April 14 and May 5, 2026. Brockman says 'not incremental' — strong GPT-6 signal. And Spud powers a super-app.
Spud's pre-training (the initial model training phase) wrapped up on March 24. OpenAI's standard post-training cycle — safety evaluation, red-teaming (external security testing), and RLHF (reinforcement learning from human feedback) — takes 3 to 6 weeks. The release window is now calculable: between April 14 and May 5, 2026.
Greg Brockman, OpenAI co-founder, dropped this in a podcast in early April: "Two years of research. Big model feel. Not an incremental improvement." It's the most direct naming signal OpenAI has sent in months.
But Spud isn't just a model. It's the brain of a super-app (unified platform) designed to follow the user from desk to car to earbuds.
April 14 → May 5: The Math OpenAI Didn't Share Publicly
Let's start with the hard fact. Sam Altman and The Information confirmed that Spud's pre-training ended on March 24, 2026. What follows is mechanical.
OpenAI's post-training cycle breaks down into three phases. Safety evaluation (internal security assessment) takes 1 to 2 weeks. External red-teaming — independent experts stress-testing the model's limits — adds 1 to 2 weeks. Final RLHF fine-tuning takes another 1 to 2. Total: 3 to 6 weeks after pre-training ends.
The math is simple. March 24 + 3 weeks = April 14. March 24 + 6 weeks = May 5.
| Phase | Date | Duration |
|---|---|---|
| Pre-training complete | March 24, 2026 | — |
| Safety evaluation | March 24 → April 7 | ~2 weeks |
| External red-teaming | April 7 → 21 | ~2 weeks |
| RLHF + final fine-tuning | April 21 → May 5 | ~2 weeks |
| Release window | April 14 → May 5 | 3 to 6 weeks |
Prediction markets agree. Polymarket (prediction betting platform) gives 78% probability of release before April 30. Manifold Markets goes to 82% before May 15. These aren't opinions — they're the best aggregators of dispersed information on this kind of topic.
Altman himself said publicly: "Things are moving faster than many expected." And confirmed a release "before Memorial Day" (U.S. holiday, late May). This isn't speculation. It's calendar engineering.
"Not an Incremental Improvement" — Why Brockman Said Everything in 4 Words
Brockman's quote deserves a word-by-word breakdown.
"Two years of research." Two years since GPT-4 (March 2023). Not since GPT-4.5 (February 2026). Brockman is talking about a full research cycle — the kind that produces a generational leap.
"Big model feel." A qualitative jump perceptible to the user. Not an improvement measurable only on benchmarks.
"Not an incremental improvement." The negation is explicit. In OpenAI's vocabulary, GPT-4.5 was the incremental improvement — and the market called it disappointing. Brockman is saying Spud will be the opposite.
The naming logic follows. GPT-4 (March 2023) → GPT-4o (May 2024) → GPT-4.5 (February 2026, disappointing) → Spud. If Spud is "not incremental," then it's not a GPT-5.5. It's a GPT-6. The cycle-breaking moment an industry disappointed by GPT-4.5 has been waiting for.
The 6 Layers of the Super-App: Spud Everywhere
The real revelation isn't the model itself. It's what OpenAI is building around it.
Spud isn't a standalone model. It's the "core intelligence layer" of a unified platform. Six layers are coming together to form a super-app.
| Layer | Product | Status | Role in the Super-App |
|---|---|---|---|
| Intelligence | Spud / GPT-6 | 🔜 April 14 → May 5 | Central brain |
| Code | Codex CLI | ✅ Updated April | Multi-agent dev |
| Memory | File Library | ✅ Q1 2026 | Persistent workspace |
| Mobility | Apple CarPlay | ✅ Announced | 100M vehicles |
| Content | TBPN acquisition | ✅ Acquired | Media layer |
| Hardware | Sweet Pea earbuds | 🔜 September 2026 | Ambient computing |
Codex CLI was updated in early April with a rebuilt multi-agent infrastructure. Spud will be the central model — every code agent runs on it.
File Library, launched in Q1 2026, brings a persistent cross-session workspace. Spud "remembers" your files and projects. Long-term memory becomes operational.
Apple CarPlay connects ChatGPT to 100 million vehicles. Spud as a permanent voice copilot in the car.
TBPN, acquired by OpenAI, adds a content creation and distribution layer. Spud generates, publishes, distributes. OpenAI becomes a media platform.
Sweet Pea, the AI earbuds expected in September 2026, target 40 to 50 million units in year one. Spud in your ears, 24/7. Not a chatbot — permanent ambient computing.
ChatGPT + Deep Research + Memory unify all existing products on Spud. One interface. One brain.
The vision: desk (Codex + File Library) → car (CarPlay) → ears (Sweet Pea) → research (Deep Research) → content (TBPN). It's the same logic as Apple with iOS — a platform that follows the user across every dimension of their life. Except OpenAI wants to move faster and without the hardware. For now.
$730 Billion and Spud Must Justify Every Dollar
The financial context makes Spud's release existential.
OpenAI is valued at $730 billion after a $110 billion raise. SoftBank borrowed $40 billion to participate. Every dollar of that valuation rests on the promise that the next models will justify the investment.
But GPT-4.5 didn't satisfy the market. Gemini 2.5 Pro has dominated LMArena (the community AI model ranking) since mid-March. Anthropic's Claude Mythos arrives around May 6. Gemma 4 31B hits 89% on AIME under open-source Apache 2.0.
| Model | Lab | Window | Signal |
|---|---|---|---|
| Gemini 2.5 Pro | ✅ Available | #1 LMArena | |
| Gemma 4 31B | ✅ Available | AIME 89%, Apache 2.0 | |
| Spud / GPT-6 | OpenAI | 🔜 April 14 → May 5 | "Not incremental" |
| Claude Mythos | Anthropic | 🔜 ~May 6 | "Most powerful ever" |
| Grok 4 | xAI | 🔜 April-May | Strong reasoning |
In this context, a GPT-5.5 would be a perception disaster. GPT-6 is the only credible answer to $730 billion in valuation. This is no longer a product question. It's a narrative survival question for OpenAI facing the alliance of competing labs.
What Spud Probably Won't Be
Let's be honest about what we don't know yet.
Pricing. Will GPT-6 cost more than GPT-4.5? Likely. Higher compute means higher pricing. The question is how much.
Multimodality. Native audio like GPT-4o? Improved vision? Nothing confirmed.
Context window. Will the 128K tokens be maintained? Extended? Unknown.
API. Available on day one or delayed as usual? OpenAI's track record isn't reassuring.
Latency. GPT-4o probably remains faster for everyday use. Spud targets frontier reasoning — like o3, but as a full-size flagship model.
And the super-app will take months to build out. Sweet Pea doesn't arrive until September. CarPlay is still rolling out. The vision is clear. The execution will be gradual. Check the full release window analysis and the Q2 2026 AI model calendar to track the evolution.
In summary:
- Spud (OpenAI's next flagship model) drops between April 14 and May 5, 2026 — calculated from pre-training completion on March 24 + standard 3-to-6-week post-training cycle
- Polymarket gives 78% probability before April 30; Greg Brockman: "not an incremental improvement" → strong GPT-6 naming signal, not GPT-5.5
- Spud is the "core intelligence layer" of an OpenAI super-app merging ChatGPT, Codex, File Library, CarPlay, TBPN, and Sweet Pea earbuds into an ambient computing platform
- Context: GPT-4.5 deemed disappointing + Gemini 2.5 Pro #1 on LMArena + Anthropic's Claude Mythos on May 6 → OpenAI must hit hard to justify $730 billion in valuation
- Open questions: final naming (GPT-5.5 vs GPT-6), pricing, latency, context window, native multimodality
Spud isn't OpenAI's next model. It's OpenAI's answer to everything that has happened since GPT-4 in 2023: two years of research compressed into a single model, the backbone of a platform that wants to follow the user through every moment of their day, and the justification for $730 billion in trust placed in a company that had disappointed with GPT-4.5. Brockman said "not incremental." In 6 to 27 days, we'll know if he was right.
