What is Google Gemini 4 Argon?
Google’s new flagship AI model, announced on 30 September 2026
Not the chemical element. For the noble gas, see what is argon?
Gemini 4 is the fourth generation of Google’s Gemini artificial intelligence models, announced on 30 September 2026. Its first model, called Argon, is larger than Google’s previous Pro models and, on benchmarks Google published, leads OpenAI’s GPT-6 Astra and Anthropic’s Claude models on most tests while trailing on several coding ones. It is not generally available yet: it is going first to cybersecurity partners, with paid API customers and Google AI Ultra subscribers promised access “as soon as possible”.
About the sources. Wikipedia had not yet covered Gemini 4 when this was written. The background on Gemini comes from Wikipedia; everything about Gemini 4 itself comes from the news reports listed under Sources, and benchmark figures are Google’s own unless stated. We will update this page as more is published.
Google spent much of 2026 fielding questions about whether it had fallen behind OpenAI and Anthropic. On 30 September it answered with Gemini 4, and a model called Argon that it describes as its most powerful yet. This is what was announced, what the numbers do and don’t show, and what is still unknown.
- Made by
- Google DeepMind
- Announced
- 30 September 2026
- First model
- Gemini 4 Argon
- Available now to
- Partners in Google’s Fairwind security programme
- Coming next to
- Paid API customers and Google AI Ultra subscribers
- Output limit
- 1 million tokens
- Introductory API price
- $2 per million input tokens, $10 per million output tokens
- Previous generation
- Gemini 3
What is Gemini?
Gemini is a family of multimodal large language models developed by Google DeepMind, first announced on 6 December 2023 as the successor to LaMDA and PaLM 2. Multimodal means the models were designed from the start to work with text, images, audio, video and computer code together. Gemini powers Google’s chatbot of the same name, and its name refers both to the Latin for “twins”, marking the merger of Google Brain and DeepMind, and to NASA’s Project Gemini.
Gemini 4 is the fourth generation. Its first model, Argon, is larger than Google’s previous line of “Pro” models, and Google says it is built “to sustain deep reasoning across complex, long-horizon workflows”.
Why is it called Argon?
None of the launch coverage explains the name. It is shared with the chemical element argon, a noble gas whose name comes from the Greek for “lazy” because it reacts with almost nothing, though nothing suggests Google intended a link.
What can Gemini 4 Argon do?
Google pitches Argon at long, demanding work: software engineering and coding, research, finance, writing and business automation. It can read images, charts and long videos. Its output limit of 1 million tokens, roughly the length of several novels, is several times the 128,000 tokens of OpenAI’s GPT-6 Astra.
The headline capability is security. Google says Argon can “autonomously find, validate, and patch critical software vulnerabilities”. Google’s own staff already use it for debugging and moving codebases between languages; VentureBeat reports that one such project rewrote a video decoder’s low-level code in the memory-safe language Rust and made it run 2.7 times faster.
How does Gemini 4 compare with GPT-6 and Claude?
On Google’s figures, Argon leads. VentureBeat counted 18 benchmarks in Google’s release: Argon led outright on 12, OpenAI’s GPT-6 Astra on three and Anthropic’s Claude Opus 5.5 on two, and Argon and GPT-6 Astra tied on the last. The benchmarking company Vals AI ranked Argon first on its index of AI models, according to TechCrunch.
It does not lead everywhere. Reuters noted that Argon trailed on two of the four coding benchmarks Google included, and the gaps there are large: 10.5 points behind GPT-6 Astra on FrontierSWE v2 and 9 points behind Claude Opus 5.5 on Terminal-bench 4.0. These are numbers chosen and reported by Google; independent testing will take longer.
| Benchmark | Gemini 4 Argon | GPT-6 Astra | Claude Opus 5.5 |
|---|---|---|---|
| GraphWalks | 84.2% | 71.8% | 66.8% |
| LVBench (long video) | 91.7% | 87.5% | 83.7% |
| Vals Finance Agent v2 | 65.4% | 53.5% | 58.6% |
| AutomationBench | 51.3% | 41.4% | 42.5% |
| DeepSWE v1.1 (coding) | 77.9% | 74.1% | 74.2% |
| FrontierSWE v2 (coding) | 55.0% | 65.5% | — |
| Terminal-bench 4.0 (coding) | 57.4% | — | 66.4% |
“—” means VentureBeat did not report that model’s score for the benchmark.
When can I use Gemini 4?
Not yet, unless you work in cyber defence. Argon is rolling out first to “trusted cyber defenders” through Google’s Fairwind Program, and Google is taking part in the U.S. government’s voluntary process for reviewing models before release. Google says broad availability will follow “as soon as possible”, starting with paying API customers and subscribers to its Google AI Ultra plan. It has not given a date.
How much does Gemini 4 cost?
For developers, Google set an introductory price of $2 per million input tokens and $10 per million output tokens, according to VentureBeat and Engadget. VentureBeat reports that the price will double to $4 and $20 after the introductory period, and that even the introductory rate is a fifth of GPT-6 Astra’s listed price. Consumer access is expected through the Google AI Ultra subscription.
Why did Gemini 4 take so long?
Reuters reported that the launch came after months of delays. A Gemini 3.5 Pro model that had been expected in June 2026 was cancelled, while Google faced questions about whether it was falling behind OpenAI and Anthropic.
What we don’t know yet
When ordinary users and most developers will get access; how Argon performs on independent tests rather than Google’s own; which other models will join the Gemini 4 family; and what the full safety evaluation found. We will update this explainer as Google and independent testers publish more.
Timeline
- 6 Dec 2023Google announces Gemini 1.0.
- Feb 2024Gemini 1.5 arrives with a one-million-token context window.
- Dec 2024Gemini 2.0 Flash is released as an experimental model.
- Nov 2025Google DeepMind releases Gemini 3 Pro Image, known as Nano Banana Pro.
- Jun 2026A planned Gemini 3.5 Pro release does not happen; Reuters later reports it was cancelled.
- 30 Sep 2026Google announces Gemini 4 and its first model, Argon.
Sources
- TechCrunch: Google releases Gemini 4 Argon, called its most powerful model yet (30 Sep 2026).
- VentureBeat: Google unveils Gemini 4 Argon, retaking benchmark lead over OpenAI and Anthropic, but in limited release.
- Reuters, via The Korea Times: Google announces Gemini 4 flagship AI model after months of delays (1 Oct 2026).
- Engadget: Google’s first Gemini 4 model is “Argon”.
- Wikipedia: Gemini (language model), background and earlier versions.
Quotes are as reported in these sources. Found a mistake, or something out of date? Write to info@wikidrift.com.
Related
Keep drifting
Ask WikiDrift, and get an answer written from Wikipedia with every claim linked to its source.
More in Science.