All We Know About Google Gemini 4: Release Date, Rumors & What to Expect

Updated October 13, 2026: Gemini 4 Argon is now public, two weeks after Google announced it on September 30, and every section below reflects the release. Our Gemini 4 Argon guide covers its benchmarks, price and rollout in full.

Gemini 4 is out. Google announced Gemini 4 Argon, the first Gemini 4 model, on September 30, 2026, and after two weeks with vetted cyber defenders only, opened it to paying users on October 13, 2026. The road there began on July 21, 2026, when Google wrote, in the announcement of a much smaller model, that “we have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress”. That was the first time Google confirmed the name in its own words.

Whether you can use it now depends on what you pay for. Argon is in the Gemini app and AI Mode in Search for Google AI Pro and Ultra subscribers, and on the paid Gemini API, but not on the free plan or Google AI Plus. This guide separates what Google has actually said from what the internet invented along the way. In it we answer when Google Gemini 4 was released, what it can do, how the rumors turned out, and which AI models to use if Argon is not on your plan.

The Key Takeaways

• Out Now, Not for Free: Google announced Gemini 4 Argon on September 30, 2026 and released it on October 13, 2026. As of that date it is on Google AI Pro and Ultra in the Gemini app and on the paid API as gemini-4-argon, but not on the free plan or Google AI Plus.
• Price and Scores: Argon costs an introductory $2 in / $10 out per million tokens until November 30, 2026, then $4 / $20. Artificial Analysis scores it 53 on its Intelligence Index v4.3.2, against 41 for Gemini 3.8 Flash, still the newest Flash. Gemini 3.5 Pro, promised for June, is still unreleased.
• Bigger, Not Just Faster: On the July 22, 2026 earnings call, Sundar Pichai said the next leap needs “much larger base models” and named coding and agentic coding as the gap Google has to close.
• Use What Works: If Argon is not on your plan, you can cover most of its workflows with Gemini 3.8 Flash on the API, or Gemini 3.1 Pro and Gemini Omni Flash in Fello AI.

Current status of Google Gemini 4

Gemini 4 is now a public model. Google announced Gemini 4 Argon on September 30, 2026 and first gave it to “a set of trusted cyber defenders” through its Fairwind Program. On October 13, 2026, the Gemini release notes announced that it “is rolling out globally for Google AI Pro & Ultra subscribers in the Gemini app”, and Google opened it on the paid Gemini API the same day. Google had said in September that Ultra subscribers and paid API customers would come first, but Pro got it at the same time as Ultra.

As of October 13, 2026, the API model catalog lists gemini-4-argon as a stable model, and Google has published a model card that gives a March 2026 knowledge cutoff. Google AI Ultra subscribers also get Argon in Google Antigravity, AI Studio and Android Studio, and business customers get it through Gemini for business. Argon is the only Gemini 4 model so far, and Google says only that it looks forward “to bringing you the rest of the Gemini 4 family”. Below it, the Gemini 3.8 generation is the newest: gemini-3.8-live and gemini-3.8-live-extended-thinking from September 15, the 3.8 Flash TTS voice models from September 23, and Gemini 3.8 Flash as the newest Flash, with Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, the Gemini Omni series, Gemini 3.1 Pro still carrying its -preview suffix, and the Gemini 2.5 stable versions alongside it.

What changed recently (Official Timeline)

To understand where we are, look at the recent official release cadence from Google.

Official announcements vs social rumors

The internet is full of speculation. Even now that Gemini 4 has been released, posts about the next Gemini 4 model or about Argon reaching the free plan are often predictions rather than facts. Google typically releases new models by updating their technical documentation and publishing a blog post on The Keyword.

It is important to understand the difference between a “model update” and a “new generation.” Google frequently pushes updates to existing models without changing the major version number, and the recent jump from Gemini 3 Pro to Gemini 3.1 Pro to Gemini 3.5 Flash is a perfect example. Google naming the model in a blog post, as it did on July 21, was a genuine signal, but a signal about the roadmap and not about availability. When the generation did arrive on September 30, it came with a post on The Keyword and announcements from Google DeepMind’s own accounts, but not with a public model ID, because the first release went to a closed group. The public release on October 13 is the one that brought the API model ID, the model card and a dated entry in the Gemini release notes.

How to check for the release

If you want to know is Gemini 4 released for real, do not rely on trending hashtags. To confirm Gemini 4 availability, rely on official sources. Consumers should watch the dated Gemini release notes, developers should check the API model catalog and changelog for new model IDs, and enterprise users should monitor the official enterprise model lists and the Google AI plans page.

Expected features and capabilities

Note: Below is inference based on Google’s public direction in agents, creation, and computer use (e.g., Project Mariner (agent prototype)) and the current Gemini roadmap signals, not a leak.

We wrote these grounded expectations about Gemini 4 features before Argon was announced, and we have kept them, with each one checked against what Google published on September 30. The 3.5, 3.6, 3.7 and 3.8 Flash releases raised the bar with frontier intelligence at Flash speed, and Gemini Omni introduced native video generation. The next step will likely focus on how the AI acts and reasons over longer tasks, moving from “chatbots” to true “agents.” Pichai pointed the same way on the July 22, 2026 earnings call, calling coding and agentic coding an area where Google has “acknowledged we need to improve” and saying the next frontier depends on “much larger base models”.

Advances in reasoning and context

Users are hoping for a more powerful Gemini 4 context window. Current Gemini API models, 3.8 and 3.7 and 3.6 and 3.5 and 3.1 and 2.5 alike, generally support 1,048,576 input tokens (≈1M). Worth knowing before you plan around that number, Gemini 3.6 Flash retrieves only 54.0% on GDM-MRCR v2 at the full 1M mark, so a huge window is not the same as reliable recall across it. Earlier Gemini 1.5 Pro releases supported up to a 2M token context window historically, but today’s competitive focus has shifted to reasoning and tooling. We expect major improvements in deeper “System 2” thinking where the model takes time to reason before answering, better reliability with fewer hallucinations for enterprise use, and richer Computer Use that builds on the Gemini 2.5 Computer Use: automate your browser tools and the agentic gains already in the 3.7 and 3.8 Flash releases. The September 15 Live models are the first public look at that direction: Gemini 3.8 Live Extended Thinking reasons and speaks at the same time, narrating its progress through multi-step background tasks instead of going quiet while it works. How it landed: Google calls Argon “built to sustain deep reasoning across complex, long-horizon workflows”. The window did not grow: Argon has 1,048,576 tokens, but that limit now covers input and output together, and with the API’s new long-decode feature a single response can run to about 1 million tokens, against 64,000 on earlier Gemini models.

Multimodal upgrades

We also expect Gemini 4 multimodal capabilities to push past today’s baseline. Gemini Omni Flash already accepts image, audio, video, and text input and generates editable video, and Google has since added a cybersecurity-tuned Gemini 3.5 Flash Cyber, so Gemini 4 will likely extend toward longer-form video, real-time streaming, and tighter integration with agents. Two September releases already move that way: agentic video understanding, added on September 1, lets Gemini search and inspect only the parts of a video it needs instead of ingesting every frame, and Gemini 3.8 Live reads visual input in near real time while it is still talking. Imagine showing the AI a video of a broken appliance and asking it how to fix it. A future Gemini 4 agent could watch the video, identify the part, and browse the web to find the replacement without prompting. How it landed: Google pitches Argon at coding, knowledge work, cybersecurity defense and creative writing rather than at media creation, but its own benchmarks show it reading long video well, with a state-of-the-art 91.7% on LVBench.

Rumors leaks and speculation

It is crucial to practice “rumor hygiene” when reading about a Gemini 4 leak. Not all sources are created equal.

Rumor Log: Confirmed vs. Unverified

Note on Project Mariner: This is evidence of Google’s agent direction (computer use), not Gemini 4 confirmation.

The newest Gemini 4 rumor is about a model called Carbon. On October 9, 2026, Business Insider reported, citing internal documents and screenshots, that Google staff are testing Carbon, a newer Gemini 4 version, in Jetski, Google’s internal coding platform. One employee said it “feels like Opus 5.5” for coding, but added that they would need to test it more. According to the same report, Google has tested Gemini 4 models internally named Argon, Barium and Carbon, and an internal model called “Barium-B” was chosen to become the public Argon. Business Insider could not determine whether Carbon will ship as an update under the Argon name or as a separate Gemini 4 model, and Google is not guaranteed to release it at all. Until Carbon shows up in a benchmark or an API catalog, treat the comparison with Claude Opus 5.5 as one employee’s first impression.

Pricing and developer access

Money matters. Google has announced Gemini 4 pricing for the API: Argon costs an introductory $2 per million input tokens and $10 per million output tokens, with cached input 95% off, until November 30, 2026, and $4 and $20 after that. API access needs the paid tier. Which consumer plans get it is a separate question, and there the naming has already evolved.

Subscriptions are now described via Google AI plan tiers, and the current ladder runs Free at $0, Google AI Plus at $4.99, Google AI Pro at $19.99, and Google AI Ultra from $99.99 with a $199.99 tier on top for 20x the Pro limits. As of October 23, 2026, Google’s plans page reads “Access to our Flash-Lite model” for the free tier and “Access to our Flash model” for Plus, with no version numbers, and lists “Access to Gemini 4 Argon” on both Google AI Pro and Ultra, next to Google’s Pro and Flash models. That makes Pro the cheapest way to use Argon in the Gemini app. Paid-first rollouts were already the pattern on the Flash line: in the Gemini app, Gemini 3.7 Flash reached you only through Spark, which needs AI Pro or Ultra, and Google excluded the European Economic Area, the United Kingdom, Switzerland and Nigeria from that rollout. You can also learn how to claim free Google AI Pro for students, which now includes Argon, and our Gemini pricing guide and wider AI pricing comparison cover the plans in detail.

Google has not said whether Argon will ever reach Google AI Plus or the free tier. Its multi-tiered approach keeps a Flash-class model for casual users and saves the frontier model for paying ones, and the shift from the old “AI Premium” to today’s “Pro” and “Ultra” levels points to a more specialized future for Google’s subscription ecosystem.

Best alternatives to use today

If Argon is not on your plan, or you want something cheaper for everyday work, the rest of the lineup covers most jobs. To understand what is already possible, see our Gemini 3.5 review of what Google launched at I/O 2026 and our explainer on Gemini 2.5 vs Gemini 3.

Note: Use *-latest aliases carefully, as they have automatically switched to newer Gemini models with each release.

Real workflows you can build now

You don’t need Gemini 4 to start high-value automation. You can already chat with PDFs using AI to analyze large documents or use our 7 best Gemini prompts for YouTube scripts to accelerate content creation.

For those who want a seamless experience, the Gemini AI desktop client for macOS is the best way to integrate these models into your daily workflow.

Stable (GA) models have published guidance on model versions, and preview models can change or be deprecated quickly. If you are building a business, stick to the Gemini 4 Argon, Gemini 2.5 Stable, Gemini 3.5 Flash, Gemini 3.6 Flash, Gemini 3.7 Flash or Gemini 3.8 Flash IDs, all of which Google lists as stable, and treat anything carrying a -preview suffix as changeable.

Conclusion

The honest summary is that Google Gemini 4 has gone from roadmap to release in under three months. Google named it and confirmed pre-training on July 21, 2026, alongside the Gemini 3.6 Flash launch, said in late September that it had entered post-training, announced Gemini 4 Argon on September 30 for Fairwind partners only, and released it to Google AI Pro, Ultra and the paid API on October 13. The free plan and Google AI Plus do not get it, and if they ever do, Google’s track record says not to bet on a date: Gemini 3.5 Pro was promised for June 2026 and has still not shipped. If Argon is on your plan, test it on your own work; if not, Gemini 3.8 Flash, the 3.8 Live voice models and the Gemini Omni creation models cover most jobs. Our Gemini 4 Argon guide has the benchmarks and price in full.

Next Step: You don’t have to wait to upgrade your workflow. Getting started with Fello AI is the best way to try the latest available Gemini models today, side by side with GPT and Claude in one app.

FAQ