A user opens the Gemini app on October 1 and looks for Argon. The model is not there. Google has announced Gemini 4 Argon, published its price and displayed benchmark charts, but public access has not started.

Google introduced the model on September 30, 2026. The first deployment covers a small group of trusted cybersecurity defenders. Google says paid API customers and Google AI Ultra subscribers will receive access later. It has not provided a calendar date for that broader release.

The distinction matters. Gemini 4 Argon is official. General availability is not. This report separates Google’s confirmed details from benchmark claims, anonymous reports and search speculation.

The timing also follows a larger cost trend. The 2025 Stanford AI Index reported that the cost of querying a model at roughly GPT 3.5 performance fell from $20 to $0.07 per million tokens between November 2022 and October 2024. That was a reduction of more than 280 times. Frontier models now compete on price, task duration and reliability, not only short benchmark answers.

What is Google Gemini 4 Argon?

Argon is Google’s new frontier Gemini AI model

Gemini 4 Argon is Google DeepMind’s next frontier model for complex work that continues across many steps. Google designed it for software engineering, financial research, legal work, cybersecurity and multimodal analysis.

The name has two parts. Gemini 4 identifies the model generation. Argon identifies this flagship model. Search phrases such as “Argon Gemini,” “Gemini Google,” “AI Gemini” and “Gemini 4.0” usually refer to the same announcement.

Google describes Argon as a model for long horizon workflows. That means the model can keep working through a large coding task, a collection of financial documents or a security investigation instead of answering one isolated prompt.

Google DeepMind product head Tulsee Doshi told Axios, “Argon is a well-rounded model that has frontier capabilities across several domains.” That description comes from Google. Independent public testing remains limited because access is restricted.

Follow Ufaq’s AI coverage and Technology section for updates as Google expands access.

What is new in Gemini 4 Argon?

A one million token output limit

Google says Argon can produce up to one million output tokens in one trajectory. The previous limit was 64,000 tokens. This is an output limit, not a confirmed one million token input context window.

The extra output space gives an agent room to inspect files, use tools, test changes and revise its work. It does not guarantee that every long response stays accurate. Long tasks create more points where an error can enter the process.

Coding, research and cybersecurity work

Google reports several internal uses. Argon helped quantum researchers improve a published optimization baseline by 40 percent. A group of Argon agents found memory changes that freed more than 300 TiB across Google’s data centers, with estimated total savings between 500 TiB and 1 PiB.

Google also used Argon agents on large code migrations. The work ranged from smaller core libraries to more than 800,000 lines in the Fuchsia Zircon kernel. In another project, agents replaced 32,000 lines of SIMD code in a Rust video decoder. Google says the result ran 2.7 times faster than the earlier Rust port while producing identical video output.

For cybersecurity, Google is giving selected defenders a version without the normal cyber guardrails. That version can search for, validate and patch software vulnerabilities. It is not the version Google plans to release to general consumers.

Security remains a live problem. A 2025 Google DeepMind report on indirect prompt injection said models that use tools can encounter malicious instructions inside emails, web pages and other retrieved data. Google now says Argon leads a Gray Swan prompt injection benchmark. Teams still need permissions, sandboxing, monitoring and human review around the model.

Gemini 4 Argon benchmarks and internal skepticism

Google reports strong scores across several tests

The official Gemini 4 Argon announcement lists the following results:

  • DeepSWE v1.1: 77.9 percent for long software engineering tasks.
  • AutomationBench: 51.3 percent for work across business functions.
  • LVBench: 91.7 percent for long video understanding.
  • CWE Bench v1: 68 percent for vulnerability repair, tied for first place.

These figures come from Google or benchmark partners named by Google. They do not prove that Argon will lead every task. A model can score well under one tool setup and perform differently in a private codebase, a regional language or an application with different permissions.

What the Google Gemini 4 internal skepticism report says

Axios reported that Bloomberg spoke with anonymous sources who had direct access to the project. Some sources said Gemini 4 performed well on industry benchmarks but struggled with parts of real coding work. Google disputed that account.

Both statements remain part of the record. Google points to thousands of employees using Argon for coding, research and writing. Anonymous sources described weaker results in some practical tasks. Public users cannot settle the question yet because they do not have broad access.

The correct test will use real repositories, fixed budgets, logged tool calls and repeatable acceptance criteria. A general score cannot replace that process.

Gemini 4 release date and access

Google announced Argon on September 30, 2026

The Gemini 4 release date has two answers. Google announced the model on September 30, 2026. As of October 1, 2026, it has not announced an exact date for general API or consumer access.

The first cohort consists of trusted cyber defenders in Google’s Fairwind Program. Google says it is also participating in a voluntary United States government process for access before release.

Paid users come next, but Google has not stated when

Google plans to start the broader rollout with paid API customers and Google AI Ultra subscribers. The company uses the phrase “rolling out soon.” That is not a fixed date.

An Ars Technica report reached the same practical conclusion: Google announced Argon, but most users cannot use it yet.

Do not pay a third party that claims to sell immediate Gemini 4 access without checking the provider, model identifier and billing route. Early labels and unofficial interfaces can point to a different Gemini model.

Gemini 4 Argon pricing and ChatGPT comparison

The introductory API price starts at $2 and $10

Google says Gemini 4 Argon will launch at an introductory API price of $2 per million input tokens and $10 per million output tokens. Cached input will cost 95 percent less than normal input.

After the introductory period, Google plans to charge $4 per million input tokens and $20 per million output tokens. The company has not stated when the introductory period will end. These prices apply to the planned API service. They are not the monthly Google AI Ultra subscription price.

The price trend fits the wider AI market, but cost per token does not equal cost per finished task. A long agent run can use many calls, tools and output tokens. Measure the entire workflow.

Google Gemini 4 versus ChatGPT

Searches such as “ChatGPT Gemini,” “chat Gemini” and “Gemini chat” combine product names that describe different services. ChatGPT is OpenAI’s assistant. Gemini is Google’s assistant and model family. Argon is a specific Gemini 4 model.

Google compares Argon with current OpenAI and Anthropic models in its charts. Those comparisons use selected benchmarks and model settings. If you need coding, test both systems on the same repository. If you need research, test source quality and citation accuracy. If you need an agent, test permissions, recovery behavior and total cost.

For background on OpenAI’s competing model family, read Ufaq’s ChatGPT model overview. Our Guides section covers practical setup and testing.

The Gemini announcement also placed “GOOG stock” among related searches. A product release can affect market attention, but it does not establish a reliable direction for Alphabet shares. This article covers the model, not investment advice.

Google Gemini 4 Argon FAQ

Common questions about Gemini AI and Argon

What is Gemini 4 Argon?
Gemini 4 Argon is Google DeepMind’s new frontier AI model for long coding, enterprise research, multimodal work and cybersecurity tasks.

Is Gemini 4 Argon available now?
Only a limited group of trusted cybersecurity defenders has access. Google plans to serve paid API customers and Google AI Ultra subscribers next.

What is the Gemini 4 release date?
Google announced Gemini 4 Argon on September 30, 2026. It has not provided an exact date for broad public access.

How much will Gemini 4 Argon cost?
The introductory API price is $2 per million input tokens and $10 per million output tokens. Google plans to raise those rates to $4 and $20 after the introductory period.

Does Argon have a one million token context window?
Google confirmed a one million token output limit. Its announcement did not describe that figure as the input context window.

Is Gemini Argon better than ChatGPT?
No single result answers that question. Google reports strong benchmark scores for Argon, but broad independent testing has not started. Compare both systems on your own tasks, budgets and safety requirements.