GPT-5.7 Release Date: Not Released Yet — Leaks, Rumors, and What to Use Now

GPT-5.7 has not been released as of August 18, 2026 — there is no model page, no API, and no confirmed date, and every "release date" you will find in search results is a leak or a rumor, not a schedule. The company has confirmed its next frontier system exists (it publishes under the name Astra), but reporting says the final name is still undecided: it could ship as GPT-6, as a point release like GPT-5.7, or as a fourth tier next to GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna. What you can actually call today is the current flagship, GPT-5.6 Sol — live on OrcaRouter at the exact $5.00 / $30.00 per million tokens, with $0 per token added and your own key. This page is the GPT-5.7 status brief: what the leaks claim, what's confirmed, what it might cost, and how to be ready the day it ships.

If you searched for "gpt 5.7", the short version is above. The longer version is that GPT-5.7 is currently a name without a product. OpenAI has not announced a model by that name; what it has confirmed is Astra, its "next major model," via an August 1 research post about ten machine-checkable mathematical results. The Information reported on July 31 that the company is still deciding whether Astra ships under a new name (GPT-6), as a point release on the current family (GPT-5.7), or as an additional tier beside GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna. We covered Astra and the ten proofs in depth in our separate Astra leak article — this page focuses on the GPT-5.7 question specifically: is it real, when does it come out, what will it cost, and what do you use while you wait.

Is GPT-5.7 real? What's confirmed vs. what's leaked

There is exactly one thing about GPT-5.7 that is confirmed: nothing by that name. What is confirmed is that OpenAI's next frontier system exists, is internally known as Astra, and is described as a long-running, multi-agent system rather than a single-pass chatbot. Everything else about "GPT-5.7" comes in one of two grades — credible reporting or single-source rumor — and treating those two the same is how otherwise careful pages end up printing a launch date.

• Confirmed — OpenAI has stated Astra is its "next major model" and published machine-checkable Lean proofs for ten mathematical results, quoting the compute cost at GPT-5.6 Sol rates. No model card, no release date, no price, and no benchmark score were part of that announcement.

• Credible reporting — The Information (July 31) reports the final product name is undecided and the family is already in testing; a separate July 28 report citing The Information and 第一财经 put GPT-5.7 in an August 2026 window. Two independent outlets on the same story, still not a schedule.

• Single-source rumor — a flash report dated August 6, citing one unnamed whistleblower, claims the latest Astra build (codename "mewfour") is a release candidate that could ship "as soon as next week," is the largest model OpenAI has trained since GPT-4.5, and could be named GPT-5.7 or GPT-6. It is one unverified claim that some coverage is already treating as a calendar date.

Our read, reconciling that flash report with the 30-day federal pre-release review that OpenAI's models have been subject to and with prediction-market pricing (the crowd still weighted an autumn release at roughly 68% by September 30 as of mid-August): a "next week" ship would require the federal submission to already exist, and nothing public supports that. No date is confirmed. If someone tells you GPT-5.7's release date, they are guessing — and this page is deliberately not going to do that.

What the leaks claim GPT-5.7 will be

Treat every item below as a rumor with a source-grade, because OpenAI has verified none of it. The recurring claims, in rough order of how often they circulate:

• A new pre-training foundation — leaks describe the Astra lineage as a "brand new" pre-trained model rather than a continuation of the GPT-5.6 weights, and as the largest OpenAI has trained since GPT-4.5.

• A context window around 1.5M+ tokens — one of the biggest claims and one of the least sourced. GPT-5.6 Sol, the current flagship, ships with ~1.05M.

• A parameter count near 10 trillion — unverified, and arguably the least decision-relevant number in the pile: parameter counts rarely drive what you can call or what you pay.

• Deeper reasoning, coding, and agentic work — multi-step reasoning, planning, multi-agent research, and code writing and debugging are the consistent themes across leaks, and they match how OpenAI itself describes Astra.

• An enterprise and security posture — large knowledge-base search, report generation, and workflow automation are repeatedly named, which matters if GPT-5.7 follows GPT-5.6's pattern of a cautious, government-reviewed rollout.

None of this is a spec sheet. It is a picture of what the next frontier model is expected to be — bigger, longer-context, more agentic — and that picture is consistent enough to plan around in broad terms, and not specific enough to build against.

What GPT-5.7 will cost — and what the flagship costs today

There is no GPT-5.7 price because there is no GPT-5.7. The only pricing signals we have are indirect: the Astra research post quoted its compute cost at GPT-5.6 Sol rates, and reporting describes the family as already in testing — neither of which says anything about final API pricing. The honest baseline for "what will it cost" is today's GPT-5.6 family, verified August 18, 2026 against the OrcaRouter directory:

• GPT-5.6 Sol — $5.00 input / $30.00 output per 1M tokens, the flagship reasoning tier. Prompt cache reads bill $0.50, the batch API runs $2.50 / $15.00, fast mode is $10 / $60, the context window is ~1.05M tokens with up to 128K output.

• GPT-5.6 Terra — $2 / $12, the balanced middle tier after OpenAI's July 30 cut.

• GPT-5.6 Luna — $0.20 / $1.20, the volume tier that became the free ChatGPT default on August 6.

• The field — Claude Opus 5 at $5 / $25, Claude Fable 5 at $10 / $50, Grok 4.6 at $2 / $6 (it matched GPT-5.6 Sol on the Artificial Analysis index at a fraction of the price), and DeepSeek V4 Pro at $0.44 / $0.88 off-peak.

When GPT-5.7 does ship with a price, the figure that reaches you matters more than the list price. On OrcaRouter the provider's rate passes through at $0 per-token markup — if OpenAI prices GPT-5.7 at X, the model page shows X and your bill is X. That is the whole pricing model: we do not add a spread, there is no prepaid balance to buy first, and you keep using the key you already have.

How to be ready for GPT-5.7 — one endpoint, your key, $0 markup

The practical question underneath "when is GPT-5.7 coming" is "how do I use it when it is here, without re-integrating my whole stack." The answer is a router that speaks the OpenAI format, because that is the format GPT-5.7 will inherit. Here is the OrcaRouter version:

• One OpenAI-compatible endpoint. api.orcarouter.ai/v1 takes the same request shape your OpenAI client already uses. When GPT-5.7 lands, it is a model-ID swap — base URL unchanged, client unchanged, no new SDK.

• 200+ models behind that endpoint today. The endpoint that will carry GPT-5.7 already carries GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, Claude Opus 5, Claude Fable 5, Grok 4.6, and the open-weight models you would route the easy traffic to. You can build the routing logic now and only the model IDs change.

• BYOK — bring your own key. Bring your existing OpenAI key and OpenAI bills you directly at its own rates; your rate limits, credits, and spend stay in your OpenAI account, with $0 per token added in between. When GPT-5.7 ships you call it through the same key.

• Guardrails before billing. The PII Shield and content policy run before a request is billed, so a blocked request returns a clean 400 and is never charged, and the Agent Firewall grades tool and MCP calls before they execute. The same guardrails sit in front of GPT-5.7 the day it arrives.

The honest boundary

This is a rumor page, so the boundaries deserve to be explicit. GPT-5.7 might never be called GPT-5.7 — the name is still undecided per The Information, and the same flash report that named it also names GPT-6 as the alternative. Every spec in the leak pile (the ~10T parameter figure, the ~1.5M context claim) is unverified; the two concrete things OpenAI has actually shown — Astra's existence and ten machine-checkable proofs — are real, and the specs are not. There is also a real risk that a next-generation frontier model ships restricted rather than broadly: GPT-5.6 Sol's public launch was held behind a federal review and a limited preview, and commentary already speculates that GPT-5.7-class models could debut U.S.-first with vetted-partner access. That would mean a "release date" for everyone and a "you can use it" date for almost nobody. Finally, the platform honesty: OrcaRouter routes and bills; OpenAI runs the weights. When GPT-5.7 exists, we will list it at the provider rate with $0 markup and your key — and until then, the flagship you can actually call is GPT-5.6 Sol.

Bottom line

GPT-5.7 is not released as of August 18, 2026. It has no confirmed date, no price, and no model card; the "ships next week" report is a single unverified claim, and the crowd-weighted expectation still points to autumn. What exists today is the GPT-5.6 family — flagship GPT-5.6 Sol at $5 / $30 per 1M, GPT-5.6 Terra at $2 / $12, GPT-5.6 Luna at $0.20 / $1.20 — served on OrcaRouter at the provider's list price with $0 per-token markup, your own key, and 200+ other models behind the same endpoint. When GPT-5.7 ships, you are a model-ID away.

Until GPT-5.7 ships, the flagship you can call today is live at zero markup on GPT-5.6 Sol on OrcaRouter — $5.00 / $30.00 per 1M, checked 2026-08-18, ready with your own key.