- Gemini 4 is not out. As of 22 September 2026 there is no model page, no API model ID, no pricing and no benchmark table.
- Google has confirmed one thing only: pre-training began on 21 July 2026, described as its “most ambitious pre-training run yet.”
- The newest shipping model is still Gemini 3.8 Flash (2 September 2026).
- “October launch”, “2M context” and the “argon” codename all trace back to posts on X. None of them carry a Google source.
Is Gemini 4 Out?
No. As of 22 September 2026, Gemini 4 has not shipped.
This is easy to check yourself. Google announces models on blog.google and publishes the technical details on the ai.google.dev models page. Neither lists Gemini 4. Right now there is:
- No announcement post
- No model card
- No API model ID
- No price list
- No official benchmark table
The newest model you can actually call today is Gemini 3.8 Flash , released on 2 September 2026.
What Google Has Actually Said
There is exactly one confirmed data point, and it is dated 21 July 2026. Google DeepMind said it had started its most ambitious pre-training run yet for Gemini 4 and was excited by the progress.
Sundar Pichai repeated the line on Alphabet’s Q2 earnings call the next day and added two details:
- Gemini 4 will be built on a significantly larger base model.
- Coding and autonomous agents are the priorities for it.
That is the entire confirmed record. No date, no price, no context window, no benchmarks, no variant names. And “pre-training has started” is not a ship date: post-training, safety evaluation and red-teaming all come after it.
So Where Did “October” Come From?
Every timeline and spec number circulating right now originates on social media. Here they are in one table:
| Date | Claim | Source | Status |
|---|---|---|---|
| 12 Aug | Gemini 4 Pro beats GPT-5.6 Sol and Claude Fable 5, 1.5M context | Anonymous X post | No eval files, no Google source |
| 14 Sep | Codename “argon”, 256K output limit, 2M context “not decided yet” | X account | Unverified |
| 17 Sep | A model labelled “gemini-3.8-flash” in Arena producing advanced SVG and 3D output, read as a Gemini 4 checkpoint | X posts | Label unconfirmed, Google and Arena silent |
| 18-21 Sep | First Pro checkpoint is out; public release in October, with Gemini 4 Flash-Lite and an updated NB2 Lite in September | An X account going by “lyra” | No Google source |
None of this has to be wrong. But none of it is confirmed either, and not one claim points back to a Google page.
The practical filter: does the number you are reading appear on blog.google or ai.google.dev? If not, it is a guess. Pages that publish a neat “Gemini 4 pricing” table for a model that does not exist are farming the search results ahead of launch.
The Timeline Logic Worth Trusting
If you want to estimate a date, use Google’s own shipping rhythm rather than leaks. The confirmed Gemini timeline:
| Model | Released |
|---|---|
| Gemini 3 Pro | 18 November 2025 |
| Gemini 3 Flash | December 2025 |
| Gemini 3.5 Flash-Lite | 21 July 2026 |
| Gemini 3.8 Flash | 2 September 2026 |
Two inferences follow, and both stay inferences:
First, if pre-training started in late July and this generation behaves like the last one, a late-year window is plausible. Gemini 3 Pro landed in November 2025, so year-end is a familiar slot for Google’s flagship releases.
Second, Pichai said on the same call that Gemini is moving to an almost monthly release cadence. The six-week gap between 3.5 Flash-Lite and 3.8 Flash supports that. Smaller interim releases, a Flash-Lite for instance, arriving before the flagship would not be a surprise.
The honest answer: the date is unknown, and October is unconfirmed.
What to Check on Launch Day
When Gemini 4 does arrive, the comparison baseline is Gemini 3.8 Flash. The confirmed numbers:
| Spec | Gemini 3.8 Flash |
|---|---|
| Released | 2 September 2026 |
| Context window | 1,000,000 input tokens |
| Output limit | 64,000 tokens |
| Input price | $0.75 per 1M tokens (promo) |
| Output price | $3.75 per 1M tokens (promo) |
| Promo ends | 31 December 2026, then $1.50 / $7.50 |
Three things are worth watching:
- Does the context window move past 1M? That is exactly why the 2M leak gets attention.
- Does the output limit grow? At 64K tokens, this is today’s most common complaint on long code generation.
- What does it cost? A significantly larger base model usually means a higher price. You can baseline your own workload now with our LLM cost calculator and compare the moment pricing lands.
If you want to see where the top of the market sits while you wait, our GPT-6 Astra write-up covers the current frontier pricing and benchmarks.
How to Track It
Skip the leak accounts and watch three places instead:
- blog.google for the announcement itself.
- The ai.google.dev models page for the API model ID, context window and output limit.
- The Google AI pricing page, the only place a dollar figure becomes real.
We will update this post on launch day with the actual pricing, context window and benchmark numbers.
Frequently Asked Questions
Q: When is Gemini 4 coming out? A: Google has not announced a date. The only confirmed fact is that pre-training started on 21 July 2026. The “October 2026” date going around comes from a post on X, not from Google.
Q: Is Gemini 4 out yet? A: No. As of 22 September 2026 there is no model page, API ID or pricing. The newest model is Gemini 3.8 Flash, released 2 September 2026.
Q: Is Gemini 4’s codename argon? A: A 14 September 2026 post on X claimed so. Google has not confirmed it.
Q: How big will the Gemini 4 context window be? A: Unannounced. Leaks mention 2M tokens while admitting it is “not decided yet.” For reference, Gemini 3.8 Flash offers 1M input tokens and a 64K output limit.
Q: How much will Gemini 4 cost? A: Unannounced. The baseline is Gemini 3.8 Flash at $0.75 input and $3.75 output per 1M tokens, a promotional price that ends on 31 December 2026.