Gemini 4 tracker: Google said one sentence, and here is what it means
On 2026-07-21 Google wrote on its official blog: "We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress." The next day, 2026-07-22, Sundar Pichai repeated the same sentence on Alphabet's Q2 2026 earnings call, adding "we are seeing at the frontier". Two independent official channels confirm the same thing: pre-training for Gemini 4 has begun. Beyond the name, Google has published nothing - no date, no price, no API model id, not one capability number. This page puts the official wording, what it actually implies, and the status of the model queued ahead of Gemini 4 side by side.
Updated 2026-09-10
Four things that are settled
First official mention of Gemini 4
Official wording: "We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress." It appears in the Gemini 3.6 Flash launch post, the first time Google named the project publicly.
Repeated by the CEO on an earnings call
Sundar Pichai said the same sentence on Alphabet's Q2 2026 earnings call, ending with "we are seeing at the frontier". The same phrasing in an earnings setting means it is reviewed, official language rather than an aside.
The easiest sentence to misread
Google said pre-training has started. After pre-training come post-training, evaluations, safety review and staged rollout. No timing of any kind was given, so reading "started pre-training" as "about to ship" has nothing behind it.
3.5 Pro has still not shipped
The same post says Gemini 3.5 Pro is "currently testing with partners" and will be made "broadly available as soon as it's ready", and that Google looks "forward to releasing 3.5 Pro soon". From 2026-07-21 to this page's update on 2026-09-10 that is 51 days, and 3.5 Pro is still not broadly available. Gemini 4 is queued behind it.
What Google said, and what it did not
On Gemini 4 the entire public statement is one sentence: the most ambitious pre-training run yet has started. That confirms three things - the project exists, it has a name, and training has begun. The list of what it does not confirm is longer: no release date, no price, no context length, no API model id, no benchmark number, and not even that the model will still be called Gemini 4 when it ships. Google has reordered its version naming before - what followed 3.5 Pro were 3.6 Flash and 3.8 Flash rather than 4 - so version numbers cannot be used to infer release order either. The two most common misreadings of this topic are treating the start of pre-training as an imminent launch, and treating a date printed by the press as an official schedule.
Why "3.5 Pro has not shipped" is the key to timing Gemini 4
Google placed two facts in the same paragraph: 3.5 Pro is testing with partners and will be broadly available as soon as it is ready, and in parallel the team has already begun pre-training the next generation. So Gemini 4 does not replace 3.5 Pro; it is queued behind it. And 3.5 Pro's own release has slipped repeatedly - outlets reported July 2026 and then 12 August 2026, both dates passed without a release, and Google never confirmed either one. A product line whose previous version is not yet broadly available cannot yield a reliable estimate for the next generation. Any source quoting a specific date is not Google.
Timeline (all checkable)
Google launches Gemini 3.6 Flash and others; the post closes by naming Gemini 4 and saying the most ambitious pre-training run yet has begun, while 3.5 Pro is testing with partners.
Sundar Pichai repeats the same wording on Alphabet's Q2 2026 earnings call, adding "we are seeing at the frontier".
Google ships Gemini 3.8 Flash and 3.8 Flash Cyber. That announcement does not update the status of 3.5 Pro and does not mention Gemini 4 again. As of this update, the official position on Gemini 4 remains the 07-21 sentence.
Three tiers: confirmed, unpublished, and what this page will not print
Officially confirmed (quotable)
1. Pre-training for Gemini 4 has started, described by Google as its most ambitious yet. 2. That statement was officially announced through two channels: the 2026-07-21 blog post and the 2026-07-22 earnings call. 3. In the same period Gemini 3.5 Pro is testing with partners, to be made broadly available when ready. 4. As of 2026-09-10 3.5 Pro is still not broadly available. All four are verbatim in the two official sources.
Not published by Google (do not plan around it)
Release date, price, context length, output cap, API model id, capability figures, and whether the model will still be named Gemini 4 - none of these has been published and none is confirmed by Google. The two dates that circulated for 3.5 Pro (July 2026 and 12 August 2026) were reportedly sourced from financial and insider channels, both passed without a release, and Google confirmed neither; this page therefore chases no further rumoured dates and tracks only official statements. One sentence for your team: pre-training is officially confirmed to have started, no timing exists, and an unreleased 3.5 Pro is still ahead of it.
The Gemini models you can use, and the ones you cannot
Shipped Gemini models
3.6 Flash (2026-07-21), 3.8 Flash and 3.8 Flash Cyber (2026-09-02) have shipped. The published entry price for 3.8 Flash is $0.75 per 1M input tokens and $3.75 per 1M output, stated by Google as an introductory rate expiring 2026-12-31, after which it becomes $1.50 / $7.50.
3.5 Pro and Gemini 4
3.5 Pro remains in partner testing with no date for broad availability. For Gemini 4 there is a single sentence about pre-training and no specification or timing at all. Neither has a callable API model id.
What is worth doing while you wait
Move the model name into configuration so switching models is one line rather than a code change. That matters even more across vendors: the Gemini and OpenAI families differ in interface, parameter naming and tool-call shape, and collecting those differences into one adapter layer beats scattering conditionals through business logic. The second thing worth doing now is measuring a cost baseline for your real workload on a model you can already call. Without a baseline you cannot tell whether a newly released model is cheaper or more expensive for you specifically, and you end up following someone else's benchmark. Neither task depends on anything unreleased.
The honest status of Gemini on QCode
Gemini models have entries in QCode's pricing table, but we have no real Gemini calls in the last 30 days, so we make no claim here that they are available - check the /models page for live status and confirm with one small call first. What does carry real traffic on QCode over the last 30 days is Claude, GPT and several open-weight Chinese lines. If your goal is to be ready to evaluate a new model on day one, the practical preparation is to build the adapter layer and cost baseline on the line you actually use today.
Frequently asked questions
When will Gemini 4 be released?
No date has been given. The only public statement is the 2026-07-21 sentence that the most ambitious pre-training run yet has started. Pre-training is separated from release by post-training, evaluations and safety review, and Gemini 4 is queued behind a 3.5 Pro that is not yet broadly available. Any source quoting a specific date is not Google.
Pre-training has started - does that mean it is close?
That inference does not hold. Pre-training is the first stage; Google said only that it began, not how far along it is or when it ends. For comparison, OpenAI said publicly on 2026-09-08 that it has an internal model in training, and gave no date either. The industry pattern is that training status gets disclosed while schedules do not.
Will Gemini 4 replace 3.5 Pro?
Google described them side by side: 3.5 Pro is testing with partners and will be broadly available when ready, while in parallel the team has begun pre-training the next generation. That places Gemini 4 behind 3.5 Pro rather than in place of it. Whether 3.5 Pro will in fact become broadly available before Gemini 4 is not stated.
Why did 3.6 and 3.8 Flash ship instead of 4?
Gemini version numbers are not a strict release order. What followed 3.5 Pro were 3.6 Flash (2026-07-21) and 3.8 Flash (2026-09-02), both in the Flash tier. The Pro tier and the next major generation move on a different cadence, and Google has never explained the numbering scheme, so version size cannot be used to infer what ships first.
Can I use Gemini on QCode right now?
Gemini models have entries in our pricing table, but there have been no real calls in the last 30 days, so we do not claim they are available. Please treat the /models page as the live status and confirm with one small call. We would rather be conservative here than have you schedule work around a promise we have not verified.
Should I wait for Gemini 4 or pick something available now?
Waiting for a model with no date, which is itself queued behind an unreleased version, is not an executable plan. The practical move is to build your adapter layer and cost baseline on a model you can call today. Then whenever Gemini 4 arrives, and whatever it ends up being called, evaluating it means running a comparison you already have rather than starting from zero.
Sources
Quotes on this page come from two official sources: Google's blog post of 2026-07-21, "Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber", and Sundar Pichai's remarks on Alphabet's Q2 2026 earnings call, published on blog.google on 2026-07-22. The release date and pricing for Gemini 3.8 Flash come from Google's launch post of 2026-09-02. This page cites no secondary coverage and adopts no date that Google has not confirmed.
Build your baseline on a model you can call today
One key calls the GPT and Claude tiers; switching is a model id change. When a new model lands, you already have a comparison to run.
Related pages
Gemini 3.5 Pro status tracker
The model queued ahead of Gemini 4, and where it stands now.
Gemini 3.8 Flash guide
Shipped 2026-09-02, with the introductory price and its expiry explained.
Gemini 3.7 Flash guide
Specification and usage for the previous Flash tier.
This page describes a model that has not been released. Every item under "officially confirmed" can be checked verbatim in the two official sources listed above. Release date, price, API model id and capability figures remain unpublished and this page does not guess at them. We promise no launch date for any unreleased model, and the live availability of Gemini models on QCode should be read from the /models page.