OpenAI's GPT-5.5 'Spud' Is Coming: What We Know

OpenAI's next major AI model is nearly ready. Internally codenamed "Spud," the model completed pretraining around March 24, 2026, and is now undergoing safety evaluations and red-teaming before its public release. If OpenAI follows its recent cadence, you could see GPT-5.5 as early as mid-April.

Here's everything we know about what's coming — and why it matters.

A New Model Built From Scratch

This isn't an incremental update. Unlike the GPT-5.1 through 5.4 releases that refined and extended the GPT-5 base, Spud represents a completely new pretrained foundation. That means new architecture, new training data, and new scale.

Sam Altman reportedly told OpenAI employees that Spud is a "very strong model" that could "really accelerate the economy." That's a bold internal assessment, even by OpenAI's standards.

The distinction matters because fine-tuned updates (like going from GPT-5.3 to GPT-5.4) typically deliver targeted improvements in specific areas. A brand-new pretrained model can deliver step-change improvements across the board — better reasoning, fewer hallucinations, stronger multimodal understanding, and more capable agentic behavior.

What to Expect From GPT-5.5

While OpenAI hasn't published official specifications, here's what industry observers and early testers are pointing to:

Stronger reasoning with fewer errors. GPT-5.4 already reduced factual errors by 33% compared to GPT-5.2. Spud is expected to push accuracy further, with particular improvements in multi-step logical reasoning.

Better agentic capabilities. The model is designed to function more as an autonomous agent than a traditional chatbot. That means handling complex, multi-step tasks over longer timeframes — think research projects, code refactoring across large codebases, or managing multi-day workflows.

Enhanced multimodal understanding. Expect improvements in processing images, documents, and audio natively within the same conversation. GPT-5.4 introduced native computer-use capabilities (the AI can see and interact with your screen). Spud should refine and extend this.

Faster, more efficient inference. OpenAI has reportedly been optimizing Spud for improved token efficiency and response times, which would reduce costs for API users and improve the ChatGPT experience.

One early tester on X described the model as "better than GPT-5.4, but not markedly so," though they noted they were testing a pre-release checkpoint, not the final version. That assessment should be taken with caution — final optimization passes often deliver meaningful improvements.

The Naming Question

There's an open question about whether Spud ships as GPT-5.5 or GPT-6. OpenAI has said the decision depends on how significant the performance leap turns out to be over GPT-5.4.

Based on the current cadence — GPT-5.1 in November, 5.2 in December, 5.3 in February, 5.4 in April — GPT-5.5 seems more likely. But if internal benchmarks show a dramatic jump, OpenAI could use the release to reset the numbering.

For developers building on the API, the name matters less than the capabilities. What matters is whether Spud opens up new use cases that weren't practical with GPT-5.4.

When Does It Launch?

Prediction markets assign a 78% probability that GPT-5.5 launches by April 30, and over 95% by June 30. Based on OpenAI's recent pattern of 3-6 weeks between pretraining completion and public release, the most likely window is April 14 to May 5.

OpenAI has been maintaining a monthly release cadence through 2025 and into 2026, and there's institutional pressure to keep up that pace. The company recently surpassed $25 billion in annualized revenue and is reportedly taking early steps toward a public listing — a steady drumbeat of model releases supports both narratives.

What This Means for the AI Race

GPT-5.5 arrives into a fiercely competitive landscape. Claude Opus 4.6 currently leads the LMSYS Chatbot Arena. Google's Gemini 3.1 Pro tops 13 of 16 major benchmarks. DeepSeek's R2 reasoning model is expected any day. xAI's Grok 4.20 introduced a novel multi-agent architecture.

To free up compute for Spud's training and deployment, OpenAI reportedly shut down Sora, its video generation model, entirely. That's a telling prioritization — it suggests OpenAI views the language model race as existentially important and is willing to sacrifice other products to win it.

For businesses and developers, the practical question is straightforward: if you're building on the OpenAI API, GPT-5.5 should deliver meaningful improvements in reliability and capability. If you've been evaluating alternatives like Claude or Gemini, Spud's release will be an important data point for your decision.

The Bottom Line

OpenAI is about to release its most ambitious model since GPT-5. Whether it ships as 5.5 or 6.0, Spud represents a full rebuild — not a refinement — of OpenAI's foundation. Watch for benchmark results, developer reactions, and pricing announcements in the coming weeks. The AI model race just got another entrant, and this one has the industry's attention.