GPT-5.7 and GPT-6: What Is Real, Rumoured and Unverified

The sensible answer to “when is GPT-5.7 coming?” is that there is no confirmed date, because there is no official GPT-5.7 announcement. The same is true of GPT-6. That does not mean OpenAI is not working on future models. It means the public evidence does not yet support firm claims about their name, size, timing or capabilities.

If you need an OpenAI model now, choose from the documented GPT-5.6 family and evaluate it on your own workload. Our GPT-5.6 review covers the current Sol, Terra and Luna options, access and pricing.

OpenAI announced GPT-5.6 on 9 July 2026. The family has three published tiers: Sol for the hardest work, Terra for balanced everyday use and Luna for speed and lower-cost, high-volume tasks. OpenAI later reduced Terra and Luna API prices, while retaining Sol as the premium capability tier.

The official GPT-5.6 material includes capability evaluations, pricing and a system card. That matters because it lets users assess the actual product rather than infer a future one from a model-number pattern.

As of this update, OpenAI has not published equivalent official material for GPT-5.7 or GPT-6. There is no public system card, model page, API identifier, release note or benchmark release that confirms either product.

The search result shown above points to a 13 July Times of AI article titled “GPT-6, GPT-5.7 Leaks: August Launch, 10T Parameters & More”. Its snippet says the rumoured upgrades could include more context, stronger reasoning, AI agents and enterprise features.

Those are plausible themes for a future frontier model, but plausibility is not confirmation. The same broad areas are already central to OpenAI's public GPT-5.6 positioning, so they do not independently establish a new model or a new architecture.

The Reddit result is weaker evidence still. It reports a user's impression that “GPT 5.7 Sol” was downgraded and hallucinating more. Individual experience reports can be useful prompts for investigation after a release, but they cannot confirm that a named model exists, that a rollout changed or that a regression is widespread.

Parameter counts are an especially poor way to judge modern models. Even when they are published, they do not reliably predict quality, cost, latency, safety or usefulness. A “10T” claim without technical documentation also leaves basic questions unanswered: whether the number refers to total or active parameters, what architecture is being counted and how the claim was measured.

Release dates are similarly easy to overstate. Model training, evaluation, safety work and product rollout can all shift a launch. It is better to say “unconfirmed” than to turn an online estimate into a deadline that readers may rely on.

We will also avoid presenting a rumoured GPT-5.7 as evidence that GPT-6 is imminent. OpenAI may release another point update, use a different name or announce nothing on the timetable suggested by third-party coverage.

A real confirmation would be straightforward: an OpenAI announcement, a public model or API page, formal release notes, pricing, a system card or clearly attributable benchmark material. Once those appear, we will update this page with the exact model name, availability, practical strengths and caveats.

Until then, readers should be wary of pages that present an August launch, 10-trillion-parameter count or detailed performance claims as settled fact. A balanced forecast can be useful, but it must remain labelled as a forecast.

GPT-5.6 is real and documented. GPT-5.7 and GPT-6 are not publicly confirmed by OpenAI. The current rumours may prove directionally right, but there is not enough evidence to recommend waiting, budgeting or rebuilding a workflow around them.

For now, evaluate the available models, keep integrations flexible and judge any future release by its published documentation and independent testing, not a parameter-count headline.

Sources: OpenAI's GPT-5.6 announcement, OpenAI's 30 July pricing update, and OpenAI's model release notes.