GPT-6: What We Already Know And What To Expect

What was widely expected to ship as GPT-6 actually shipped on April 23, 2026, as GPT-5.5, codenamed "Spud" during development. The memory and personalization features Sam Altman teased for "GPT-6" landed in 5.5. So when people search "GPT-6" today, the model they were waiting for is here, just under a different number. Real GPT-6 is now further out. As of mid-April 2026, OpenAI hasn't published an architecture, parameter count, pricing, or launch date for it.

Here's where things stand and what to expect from the actual next-next model.

What actually shipped in mid-2026: Spud became GPT-5.5

OpenAI finished pre-training the model codenamed "Spud" on March 24, 2026 at the Stargate data centre in Abilene, Texas. On the same day, Altman publicly described the launch as "a few weeks" out. The model dropped on April 23, 2026, but as GPT-5.5, not GPT-6. It launched in three variants (standard, GPT-5.5 Thinking, GPT-5.5 Pro) and hit 88.7% on SWE-bench Verified and 82.7% on Terminal-Bench 2.0.

Why the renaming? OpenAI hasn't said publicly. The practical result is that the model people had been calling GPT-6 across X, Reddit, and prediction markets shipped as a 5.x point release. That's a positioning choice consistent with how OpenAI now frames progress around incremental capability gains rather than version-number leaps.

GPT-6 still exists as a future plan. As of April 12, 2026, OpenAI has released no architecture paper, no parameter count, no pricing sheet, and no launch date for it.

What changed in GPT-5.5 (and what GPT-6 might add)

The features Altman teased a year ago, in the August 2025 CNBC interview, landed in 5.5. He emphasized: "people want memory" and described a model that remembers tone, past questions, project details, and personal preferences across sessions. The criticism that drove the pivot was GPT-5's rollout, which users called colder and less helpful than GPT-4.

5.5 delivers a chunk of that direction. It also pushed long-context recall sharply: 74.0% on MRCR v2 at 512K to 1M token contexts, up from 36.6% in 5.4. That makes whole-codebase or whole-conversation reasoning viable in a way prior GPT-5 variants weren't. GPT-5.5 Thinking inherits the reasoning approach OpenAI piloted in o1.

What GPT-6 might add on top: deeper persistent memory with stronger encryption, better agentic loops, and personalization that survives across product surfaces. Altman has flagged that current "temporary memory" isn't encrypted, which is a real privacy concern. None of this is announced.

{{blue-cta}}

The agentic capability question

Even before Spud, pressure on OpenAI to ship agentic features was high. The current crop of agent platforms like AutoGen, LangGraph, and CrewAI has matured well past the Auto-GPT and BabyAGI prototypes from 2023. Enterprises are running agentic workflows in production today. They don't need GPT-6 to start.

For builders: design and govern your agentic workflows on whatever model you use now, and treat model upgrades as substitutions rather than re-architectures. Voiceflow's Agent Builder gives you a visual canvas and step library for orchestration. The Knowledge Base gives agents persistent context across sessions, the same direction GPT-5.5 went. Evaluations let you measure agent quality across versions so a model swap doesn't introduce regressions. SOC 2 Type 2 and PII masking handle the privacy concerns that come with persistent memory.

For deeper context on this layer, see context engineering: the discipline of dynamically assembling what the model sees on each call.

Why GPT-6 isn't on a fixed timeline

Altman has said OpenAI rarely sets high-confidence targets beyond six months out. Combined with the Spud-to-5.5 renaming, this suggests OpenAI is leaning toward more frequent point releases (5.4, 5.5, 5.6...) rather than another big-number leap. Critics like Gary Marcus have argued for a while that frontier models are hitting incremental rather than dramatic capability gains. The GPT-5.5 release supports that read.

If you're planning a stack on the assumption that GPT-6 lands in late 2026 or early 2027, plan for a range. The next big jump might come as 5.7, or as a release from Anthropic or another lab, before it's called GPT-6.

{{blue-cta}}

Bottom line

Memory, personalization, and agentic capability are not future GPT-6 features anymore. They're shipping today, primarily in GPT-5.5. GPT-6 itself is uncommitted and, based on Altman's recent posture, may end up named something else entirely.

Practical implication for builders: don't wait. Design memory-aware workflows, build governance and oversight into your agent stack, and treat model upgrades as drop-in substitutions. Whether the next model is called GPT-6 or 5.7 or something else, the architecture you build around it should be the same.