GPT-5.7:
the rumors, the codenames, and one myth
This question now has an answer. OpenAI's next generation shipped on 2026-09-03 as GPT-6 Astra, with API model id gpt-6-astra. "GPT-5.7" was only one of the candidate names during the rumour phase and never became a product. This page keeps the rumour record, including the debunked 1.5M context claim, so you can calibrate how much such leaks are worth.
OpenAI never released a model called GPT-5.7. During the rumour phase, outlets reported an internal codename Astra with the naming undecided between GPT-6 and GPT-5.7; the 2026-09-03 launch settled it as GPT-6 Astra. This page keeps the labelled rumour record and does not treat social posts as fact.
Status refreshed 2026-08-30.
Two kinds of information, kept apart
What has happened and what people say will happen are easiest to confuse when they are mixed together.
✅ Confirmed
What you can actually use today is GPT-6 Astra (shipped 2026-09-03, model id gpt-6-astra, 1,050,000 context, 128,000 output, list price $10/$50) and the three GPT-5.6 tiers. The official pricing page currently shows Sol at $4 input and $20 output per 1M tokens, noting that this promotional pricing is available at least through 2026-11-21; Terra is $2/$12 and Luna $0.20/$1.20. One QCode key calls all of them; switching is a model-id change.
⏳ Rumored (none confirmed by OpenAI)
Claims of an August ship, a new pretraining base, and ~10T still have no official landing as of 2026-08-30. Two internal codenames still circulate, plus a 1.5M context window that has been denied (below). Unlanded rumors are not a plan.
The rumors in detail
A new pretraining base
GPT-5.7 is rumored not to be a fine-tuned iteration of GPT-5.6 but to sit on a new pretraining base. If true, that would imply a larger jump than a typical point release.
unconfirmed by OpenAI
Scale and internal codenames
The training run is rumored to be around the 10T scale, and two internal codenames circulate — zinc and magnesium — with one of them rumored to target Claude Opus 5. Codenames on their own confirm no product line.
unconfirmed by OpenAI
Where the gains are said to be
During the rumour phase the community expected a clear step over the GPT-5.6 Sol of the day on agentic behaviour and reasoning. What actually shipped was GPT-6 Astra; OpenAI published no SWE-bench Verified score on either the model page or the system card, so this page still says unpublished rather than inventing one.
unconfirmed by OpenAI
About the claim that GPT-5.7 has a 1.5M context window
This one travelled widely but has been denied — there is no basis for it. OpenAI has published no context-window specification for GPT-5.7, so the correct answer is unpublished rather than any particular number. For reference: the GPT-5.6 tiers carry a 1.05M API context (400K under the Codex and LiteLLM convention).
If you find a confident context figure elsewhere, check where it came from — there is currently no verifiable official source for one.
GPT-5.7 is nowhere yet; GPT-5.6 already works
Rather than wait on a version with no date, get the work in front of you moving. All three GPT-5.6 tiers are GA with transparent official rates, and on QCode they share one key with Claude and Gemini.
What to do in the meantime
Three things you can do today, all of them more useful than refreshing rumor feeds.
Get fluent with GPT-5.6 first
Knowing what Sol, Terra and Luna are each good for, and when to move between them, is experience that still holds after 5.7 lands.
Set it up in Codex CLI
Four steps, five minutes. New releases usually just need a different model id — you will not have to redo the setup.
Work out your real costs
Official rates for all three tiers, cache billing and worked examples. Build the baseline now so you can judge whether the next version is worth it.
This page contains rumors that OpenAI has not confirmed; each is labelled as such. Official status rechecked 2026-08-30: still unreleased, no date. Rumors are not a basis for purchasing or technical decisions; we will update this page once OpenAI ships.
Frequently asked questions
When is GPT-5.7 coming out?
OpenAI has published no release date. The community rumor pointed at August; as of 2026-08-30 that window has passed and nothing shipped. This page updates only when a checkable official line appears — social posts are not a calendar.
Does GPT-5.7 have a 1.5M context window?
There is no basis for that. The claim has been denied, and OpenAI has published no context-window specification for GPT-5.7. The correct answer today is unpublished — any confident figure lacks a verifiable official source.
How much better will GPT-5.7 be than GPT-5.6?
The only confirmed part is where GPT-5.6 stands today. For 5.7 the rumors say a new pretraining base, substantially better agentic and reasoning work, and a codename said to target Claude Opus 5 — all unconfirmed, with no official benchmark numbers published.
Should I wait for GPT-5.7 or use GPT-5.6 now?
Waiting for a version with no announced date has a real cost. The three GPT-5.6 tiers are GA and span most needs, from Luna at $0.20/$1.20 to Sol at $4/$20 (the current figure on the official pricing page, which notes the promotional rate runs at least through 2026-11-21). On QCode, when a new version does ship it is usually just a model-id change — the usage patterns and cost baseline you build now carry over.
Stop waiting on the next version and finish the work in front of you
The GPT-5.6 tiers, Claude and Gemini share one QCode key and one quota. Plans from $8.57/mo; the key goes live the moment payment clears.