OpenAI "Spud" Model: What We Know About the Next ChatGPT (2026)
On March 25, 2026, Sam Altman confirmed in an internal memo that pre-training on OpenAI's next major model — codenamed "Spud" — is complete. He said it will launch within "a few weeks." This is what we know, and what it means for AI users.
OpenAI finished pre-training its next model "Spud" on March 25. Altman says release is weeks away — April 2026. Spud is natively multimodal (text + audio + images + video in one architecture) and will push OpenAI's agentic capabilities further. Sora was killed partly to free up compute for Spud. Happycapy runs on Claude (Anthropic) — it is unaffected by OpenAI's model roadmap and doesn't reset its memory when OpenAI ships a new version.
What Is Spud?
According to reporting by The Information on March 25, 2026, OpenAI CEO Sam Altman sent an internal memo confirming that the company has completed pre-training on its next major model, codenamed "Spud." Altman described it as "very strong" and said it would be ready to release within a few weeks — pointing to an April 2026 launch.
Spud is expected to be OpenAI's first natively multimodal model — meaning it processes text, audio, images, and video through a single unified architecture rather than routing requests through separate specialized models. Previous ChatGPT versions combined GPT for text, DALL-E for images, and Whisper for voice; Spud integrates these into one foundation.
Altman described Spud as a tool to "truly accelerate the economy," emphasizing agentic workflows where AI handles end-to-end tasks without human hand-holding. The organizational rename of Fidji Simo's division to "AGI Deployment" — and the merger of ChatGPT, Codex, and Atlas into a planned desktop super app — signal that Spud is positioned as the engine for a major platform shift, not just another incremental model update.
Why OpenAI Killed Sora for Spud
OpenAI shut down its Sora video generation app and API on March 25, 2026 — the same day the Spud memo leaked. The connection is direct: Sora was consuming enormous GPU resources (reportedly generating losses near $15 million per day) while Spud required massive compute for post-training. Something had to give.
The shutdown also voided a $1 billion investment and licensing agreement with Disney. The fact that OpenAI was willing to absorb that cost is a signal of how much the company is prioritizing Spud's launch. This is not a normal model iteration — it is a strategic pivot.
Spud vs Current ChatGPT — What Changes
| Feature | Spud (upcoming) | GPT-5.4 (current) |
|---|---|---|
| Natively multimodal | Yes — text, audio, image, video unified | GPT-5.4 (separate models) |
| Release timeline | April 2026 (within weeks) | GPT-5.4 live now |
| Agentic capabilities | Significantly improved (reported) | GPT-5.4 computer use |
| Real-time audio | Native (confirmed rumor) | Via separate voice mode |
| Persistent memory | Unknown — likely session-only | Session-only |
| 150+ skills ecosystem | No | No |
| Happycapy affected | No — runs on Claude | No — runs on Claude |
What Spud Doesn't Change
Spud will be a more capable ChatGPT model — but the structural limitations of ChatGPT as a platform remain. A better model does not add persistent cross-session memory. It does not give users a 150+ skill ecosystem. It does not deliver task results to your email inbox. It does not connect to your local Mac.
Every time OpenAI releases a new model, the conversation restarts the same way: what did the new model improve? The underlying platform architecture — session-isolated, tool-poor, memory-optional — stays the same.
Happycapy is built on Anthropic's Claude, which improves on its own independent roadmap. When Spud ships, Happycapy users won't notice — their agent will still have full memory of every past session, all 150+ skills available, and Mac Bridge for local automation. The platform advantage doesn't depend on which model OpenAI ships next.
Persistent memory, 150+ skills, Mac Bridge, and Claude's independent model roadmap. The best AI agent experience doesn't depend on what OpenAI does next.
Try Happycapy Free →Frequently Asked Questions
Spud is the internal codename for OpenAI's next major AI model, expected to launch in April 2026. Pre-training was completed on March 25, 2026, according to an internal memo from CEO Sam Altman reported by The Information. Spud is described as natively multimodal — handling text, audio, images, and video through a single unified architecture rather than separate models stitched together. It is expected to significantly advance OpenAI's agentic AI capabilities.
Sam Altman said Spud would be ready for release 'within a few weeks' of March 25, 2026, pointing to an April 2026 launch. OpenAI has accelerated its release schedule to free up compute from Sora's shutdown. No specific release date has been confirmed as of March 27, 2026.
Spud may be released as part of the GPT-5 family (e.g., GPT-5.5 or GPT-5.6) rather than as a separately branded GPT-6. OpenAI's naming has been inconsistent — GPT-5.4 was released in March 2026 under that label rather than GPT-6. The final product name will likely be confirmed at launch.
No. Happycapy runs on Anthropic's Claude models, not OpenAI's GPT series. Spud is an OpenAI model that powers ChatGPT. When OpenAI upgrades its model, it does not change Happycapy's capabilities. Anthropic independently improves Claude on its own roadmap. The structural advantages Happycapy has — persistent memory, 150+ skills, Mac Bridge, Capymail — exist regardless of which model OpenAI releases.