The 10T-Parameter Successor to GPT-5.6: A First Look at GPT-6

GPT-6 is OpenAI's next-generation frontier language model, built on a new pretraining base larger than the ~4T-token "Spud" foundation shared by GPT-5.5 and GPT-5.6, and it has not been officially released or dated as of July 21, 2026. The model first surfaced in community leaks in early July 2026 as the generational successor to the GPT-5.x line, positioned as OpenAI's response to Anthropic's Fable and Mythos family. OpenAI has not published a model card, spec sheet, pricing, or launch date; every number below that is not attributed to OpenAI directly comes from leaks and should be treated accordingly.

Key Takeaways

• GPT-6 is the planned generational successor to OpenAI's GPT-5.x line, sitting above GPT-5.6 (Sol / Terra / Luna), which shipped July 9, 2026.
• It is expected to run on a brand-new pretraining foundation, not a continuation of the ~4T-token Spud base used for GPT-5.5 and GPT-5.6.
• Community leakers cite a rumored 10T-parameter scale and a 1.5M+ token context window, both unconfirmed by OpenAI.
• Sam Altman publicly emphasized memory and personalization as the defining direction for the model after GPT-5, in an August 19, 2025 CNBC interview.
• No official release date. Polymarket odds as of mid-July 2026 placed the August 31 release probability at ~24% and September 30 at ~71%.
• Competitive pressure from Anthropic's Fable/Mythos family and open-weight challengers like Kimi K3 is repeatedly cited as the reason for a faster generational cadence.

What Is GPT-6?

GPT-6 is the codename used across the AI community and by OpenAI-adjacent leakers for OpenAI's next flagship model after GPT-5.6. It is not a point release inside the GPT-5 family. Multiple leak sources describe it as a fresh pretraining run on a new, significantly larger base — a generational reset rather than another 5.x increment.

The name first appeared in verifiable form in mid-2025, when OpenAI CEO Sam Altman told CNBC that GPT-6 would arrive faster than the gap between GPT-4 and GPT-5 and would center on memory and personalization. Through the first half of 2026, the community expected the "Spud" pretraining base to become GPT-6, but that model shipped as GPT-5.5 on April 23, 2026, followed by GPT-5.6 on July 9, 2026. Only after GPT-5.6's release did leakers converge on the current picture: GPT-6 is a new, larger pretrain still ahead of the public.

As of July 21, 2026, GPT-6 has not been released, previewed, or officially acknowledged with specifications. The status is best summarized in one line from a July 20 post by leaker Haider: "Even if GPT-6 has already finished training, it's unclear what 'release' means" given the current US government situation around frontier models.

GPT-6 at a Glance

Field	Value
Developer	OpenAI
Type	Frontier large language model (multimodal expected)
Predecessor	GPT-5.6 (Sol / Terra / Luna tiers)
Pretraining base	New foundation, distinct from ~4T-token "Spud" (unconfirmed size; leaks say ~10T scale)
Context window	Not yet confirmed (community leaks: 1.5M+ tokens)
Parameter count	Not yet confirmed (leak-only claim: ~10T scale)
Modality	Not yet confirmed; multimodal continuation expected
Pricing	Not yet confirmed
Availability	Not yet released; no announced date
License	Proprietary (weights not released)
Model card	Not yet published

How GPT-6 Works and What Makes It Different

The single most repeated claim across leakers is that GPT-6 is not a continuation of the Spud pretraining foundation. GPT-5.5 and GPT-5.6 shared that ~4T-token base. GPT-6 is described as a new run on a much larger corpus and a bigger model — Haider put it at "possibly 2-3x larger" than Spud, and OpenAI engineer Roon was cited by watchers as indirectly signaling that GPT-6 will be a "substantial upgrade over the latest model GPT-5.6," per a July 11 post from Mark Kretschmann.

Three coined anchors show up consistently in community discussion:

• Spud Successor — the framing that GPT-6 replaces, rather than extends, the Spud base. This is what makes it a generational step and not GPT-5.7.
• Fable++ — Bindu Reddy's July 21 characterization that GPT-6 will be "considerably more efficient than Fable," Anthropic's frontier model, on very complex workloads.
• Serving Architecture Advantage — the argument that OpenAI's edge is not just the model but the infrastructure to serve it at scale, which Haider argued Anthropic has struggled with on the Fable side.

On direction, the older but on-record signal is Altman's memory framing. In August 2025 he told CNBC: "People want memory. People want product features that require us to be able to understand them." That points at persistent cross-session state, personalization, and agentic behavior as the product differentiators, though some of that direction has already surfaced in GPT-5.5 and GPT-5.6.

What You Can Do With GPT-6

Because GPT-6 is not released, the concrete use cases below are extrapolations from the direction OpenAI has publicly stated and from the capabilities GPT-5.6 already supports.

• Long-horizon coding: A larger context window (leak-claimed 1.5M+ tokens) would extend whole-repository reasoning workflows that GPT-5.6 already handles at 512K–1M. If you need production-grade coding today, the current flagship is GPT-5.6.
• Agentic workflows: More reliable multi-step tool use is a repeated leak theme. Enterprises already running agents on LangGraph, AutoGen, or CrewAI stacks are unlikely to need to rewrite their orchestration layer.
• Cross-session memory: If the memory features Altman flagged in 2025 land at generational scale in GPT-6, personalization state that survives across product surfaces becomes a first-class primitive.
• Multimodal reasoning: Continued native handling of text, images, audio, and structured tool calls is expected but not spec'd.

How GPT-6 Compares

The comparison landscape for GPT-6, based on the models it will ship alongside:

Model	Vendor	Status	Notable claim
GPT-6	OpenAI	Unreleased	New pretrain base, leak-only 10T scale, 1.5M+ context
GPT-5.6 (Sol)	OpenAI	Released July 9, 2026	Current OpenAI flagship, ~4T Spud base
Claude Fable 5	Anthropic	Released	The frontier model GPT-6 is repeatedly positioned against
Kimi K3	Moonshot	Released	Open-weights challenger closing the frontier gap

For a deeper look at the competitive frame, see our earlier analysis of the Claude Opus 5 vs GPT-5.6 matchup and the Grok 4.6 vs Kimi K3 efficiency comparison.

Availability: How to Access GPT-6

GPT-6 is not available through any channel as of July 21, 2026. There is no API endpoint, no ChatGPT tier, no waitlist, no preview program that OpenAI has confirmed publicly. When access opens, the primary channels will be OpenAI's own surfaces: the ChatGPT product for consumers and the OpenAI API for developers, with model-card documentation on the OpenAI site.

Polymarket contract prices as of mid-July 2026 imply the market's release expectations: roughly 1% by July 31, 3% by August 14, 24% by August 31, 71% by September 30, and 88% by December 31, 2026. Prediction market odds are directional signals, not forecasts.

What We Don't Know Yet

Every one of these is genuinely open — do not treat leak numbers as spec:

• Exact release date and product name: Whether the model ships as "GPT-6" or a different label is not confirmed.
• Parameter count: The ~10T scale claim comes from a single leak thread with no OpenAI confirmation.
• Context window: The 1.5M+ token figure is unverified. The effective context (how well the model uses it) would matter more than the raw ceiling.
• Pricing: No per-token rates, no tier structure.
• Benchmarks: No independent evals exist because no one outside OpenAI has the model.
• Safety and access controls: Recent US-government export dynamics around frontier models — the same context that shaped Anthropic's Mythos rollout — apply here.

Frequently Asked Questions

Is GPT-6 released yet?

GPT-6 has not been officially released as of July 21, 2026. OpenAI has not published a model card, pricing, or launch date. Community leakers point to a late-July or August 2026 window, but nothing is confirmed.

What is the difference between GPT-6 and GPT-5.6?

GPT-6 is expected to run on a new, larger pretraining base, while GPT-5.6 shares the ~4T-token Spud foundation with GPT-5.5. GPT-5.6 was the final planned model in the 5.x series before OpenAI's next generational step, according to July 2026 community leaks.

How many parameters does GPT-6 have?

The parameter count is not confirmed by OpenAI. Community leaks reference a 10T-scale training run and a foundation 2-3x larger than Spud, but these numbers are unverified and should not be treated as spec.

What is the context window of GPT-6?

The context window is not officially confirmed. Pre-release leaks claim a 1.5M+ token context window, which would be a large jump over GPT-5.6, but OpenAI has published no spec.

When will GPT-6 be released?

No official release date exists. Multiple community leakers point to late July or August 2026, and Polymarket odds sat around 24% for an August 31 release and 71% for September 30 as of mid-July 2026.

How much will GPT-6 cost?

Pricing has not been announced. Community speculation suggests GPT-6 will be more efficient than Anthropic's Fable class on complex workloads, but no per-token rates are public.

Is GPT-6 open source?

GPT-6 is not open source. OpenAI has not released weights for any GPT model since GPT-2, and there is no indication that changes with GPT-6.

What to Watch Next

Three concrete signals will move this page from leak-tracking to reference: an official OpenAI blog post or model card (this is the definitive trigger), any changelog entry in the OpenAI API that names a gpt-6 model ID, and Polymarket price movement around the September 30 contract, which the market currently reads as the most likely resolution date. This page will be amended in place as each lands.

Building similar frontier-scale chat and agent workflows? On kie.ai you can try GPT-5.6, Claude Opus 4.8, and Gemini 3.1 Pro.