Gemini 4 Argon: what it costs, how good it is and when you can use it

Google announced Gemini 4 Argon on 30 September 2026 as the first model in its Gemini 4 generation. Almost nobody outside Google can use it yet. Availability, prices and independent scores were checked on 3 October 2026.

The answer in brief

Gemini 4 Argon is Google’s most capable AI model, and you cannot use it yet. Independent tests put it level with OpenAI’s GPT-6 Astra and behind Anthropic’s Claude Opus 5.5, at a lower cost per task than either while its introductory price lasts. Do not pause any plans for it; prepare a test for when it opens.

• Availability: only trusted cyber defenders in Google’s Fairwind Program so far. Paid Gemini API customers and Google AI Ultra subscribers come next, with no date given.
• Price: $2 per million input tokens and $10 per million output tokens at introductory rates, then $4 and $20. On independent tests that came to $1.99 per task.
• Independent score: 53 on the Artificial Analysis Intelligence Index, a combined score from ten tests. GPT-6 Astra also scores 53; Claude Opus 5.5 scores 58.
• Google’s own benchmarks: the top score on 12 of 18, with the biggest leads in legal work, very long inputs and business automation. Rivals still lead several coding tests.
• Fewer confident mistakes: a 15% hallucination rate, meaning how often it states a wrong answer as fact, against 51% for GPT-6 Astra.
• The catch: it writes a lot per task, and Bloomberg reported that some Google staff found its real coding work weaker than its benchmarks suggest. Google disputes that.
• For Search: Google has not said Argon will power AI Mode or AI Overviews.

What is Gemini 4 Argon?

Gemini 4 Argon is Google’s new frontier model, meaning one of the most capable AI models available, and the first in Gemini 4, the generation after Gemini 3. Koray Kavukcuoglu, Google DeepMind’s SVP and Chief AI Architect, wrote in the launch post that it is built for long, multi-step jobs, and named four strengths: software engineering, professional work such as legal and finance, cybersecurity defence and creative writing.

The name breaks with the Pro and Flash labels of earlier Gemini generations, and Google has not announced any other Gemini 4 model. Google says thousands of its own staff already use Argon, for example to rewrite a video decoder so that it runs 2.7 times faster than an earlier Rust version. That is Google’s account, not an independent result.

Argon reads text, images, video and speech and replies in text, according to Artificial Analysis, which lists a context window of 1 million tokens. The context window is how much material the model can read in one go: on a rule of thumb from Google that 100 tokens is about 60 to 80 English words, that is roughly 600,000 to 800,000 words (AIWIZ’s arithmetic). Argon can also write unusually long answers, up to 1 million tokens in one response, against 64,000 on earlier Gemini models.

Is Gemini 4 Argon available yet?

No. Gemini 4 Argon has no public release date, and on 3 October 2026 only trusted cyber defenders in Google’s Fairwind Program could use it. Google is releasing it in stages while it takes part in the US government’s voluntary pre-release testing process. Next in line are paid API customers, meaning businesses that use Google’s developer access to build Gemini into their own tools, and Google AI Ultra subscribers.

Fairwind, launched on 2 September 2026, is aimed at government cyber authorities, critical infrastructure operators and core technology platforms. Members must limit access to their internal security, incident response or penetration testing teams. It is not a route in for a marketing department.

Argon was not yet in the Gemini API model list or on the Gemini API pricing page, and the launch post says nothing about which countries get it first. In the UK, Google AI Ultra costs from £79.99 a month on the Google AI plans page for the UK, which did not mention Argon on 3 October 2026, so do not buy Ultra for Argon until Google lists it there. A week before launch, Kavukcuoglu told The Information that Google intended to release an early version as soon as possible.

How much does Gemini 4 Argon cost?

Gemini 4 Argon will cost $2 per million input tokens and $10 per million output tokens during an introductory period, rising to $4 and $20 afterwards. Google has not said when the introductory period ends; Artificial Analysis says the discount runs for at least a month. A token is a small piece of text, about four characters. Text you send repeatedly, such as standing instructions, can be cached and reused at 95% off, or $0.10 per million tokens. All prices here are US list prices.

Cost per job is the more useful figure. On Artificial Analysis’s tests, Argon cost $1.99 per task at introductory prices, against $3.26 for GPT-6 Astra and $5.98 for Claude Opus 5.5. At standard prices it would cost $3.98 per task, about 1.2 times GPT-6 Astra. Sources: Google for Argon, the Gemini API pricing page for Gemini 3.8 Flash, and our GPT-6.1 Sol and Claude Opus 5.5 guides for the OpenAI and Anthropic rates.

Argon gets there through cheap tokens, not short answers. It averaged 62,000 output tokens per task, against 27,000 for GPT-6 Astra. Set a length limit in any automated workflow: one response that runs to the full million output tokens would cost $10 at introductory prices and $20 at standard prices (AIWIZ’s arithmetic from Google’s prices).

How does Gemini 4 Argon compare with GPT-6 Astra and Claude Opus 5.5?

On independent tests, Gemini 4 Argon is level with GPT-6 Astra and behind Claude Opus 5.5, but costs less per task than either while its introductory price lasts. Artificial Analysis ran Argon at its most thorough setting through Intelligence Index v4.3.2, a combined score from ten tests covering agentic work (carrying out multi-step tasks on its own), coding, reasoning, knowledge and long documents. AIWIZ has not been able to test Argon itself, so every score in this post is Google’s or Artificial Analysis’s.

Each model ran at its highest setting: high reasoning for Argon, max effort for the others. Argon does not beat Opus 5.5 on this test. It joins the group just behind it, alongside GPT-6 Astra and Claude Fable 5.1, at a lower cost per task than either. GPT-6.1 Sol is the awkward comparison: one point lower, at about a third of Argon’s cost per task. For Google it is still a big jump: 23 points above Gemini 3.1 Pro Preview, its previous non-Flash model, and 12 above Gemini 3.8 Flash.

Argon makes fewer confident mistakes than its rivals. On AA-Omniscience, a knowledge test that penalises wrong answers, its hallucination rate is 15%, the lowest of any model scoring 45 or more on the index, against 51% for GPT-6 Astra and 54% for GPT-6.1 Sol. It also answers fewer questions correctly than Astra, 50% against 63%, which suggests it is more willing to say it does not know. For factual content, that trade is usually worth having.

Coding is the weak spot. Artificial Analysis puts Argon first on its version of AutomationBench, a test of multi-step business tasks, at 78%, but on Terminal-Bench 4.0, a coding test, it scores 57%, behind Claude Sonnet 5.5, Claude Opus 5.5 and GPT-6 Astra. Bloomberg reported on 30 September that some Google employees who had used Argon found it weaker on real coding work than its benchmarks suggest. Google disputes that: Tulsee Doshi, who leads Gemini products at Google DeepMind, said staff had tested the model heavily in recent weeks and that many rely on it for their hardest coding and research problems.

These Artificial Analysis figures are from index v4.3.2, checked 3 October 2026, the same version used in our GPT-6.1 Sol guide.

What benchmarks has Google published for Gemini 4 Argon?

Google published results on 18 benchmarks comparing Gemini 4 Argon with GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5. Argon has the top score on 12 and ties GPT-6 Astra on one, the CWE-bench v1 security test, at 68%; GPT-6 Astra leads on three and Claude Opus 5.5 on two. Google ran Argon at its highest setting, while most rival scores are the providers’ own published figures, according to its evaluation methodology. Treat the table as Google’s case; the full set is on Google DeepMind‘s Gemini page.

Argon’s biggest leads are in professional work, where its Harvey legal score is almost three times the next model’s, in reasoning over very long inputs and in business automation. It also leads on long video. Coding is mixed: it leads on DeepSWE but trails by around ten points on FrontierSWE and Terminal-Bench 4.0.

Why did Google skip Gemini 3.5 Pro?

Google has not given an official reason, but its leaders have said it needed a larger Gemini 4 model to compete at the frontier and chose to focus on its cheaper Flash models in the meantime. Sundar Pichai announced Gemini 3.5 Pro at Google I/O on 19 May 2026 and promised it the following month; it missed June, and a rumoured July date passed too, as our Gemini 3.5 Pro timeline records.

On Alphabet’s earnings call on 22 July 2026, Pichai described 3.5 Pro as in testing, then told analysts that the next generation of frontier models needs much larger base models, and that Google would “need Gemini 4 as a larger base model” to compete there. In September, Kavukcuoglu said Google had taken a step back from 3.5 Pro to focus on its Flash models. Neither said 3.5 Pro was cancelled, and on 3 October 2026 Google DeepMind’s Gemini Pro page still showed a “3.5 Pro coming soon” label. In effect, Argon is the top-tier model Google shipped instead.

How is Google handling safety and security?

Google is holding Gemini 4 Argon back from public release while it tightens safeguards against misuse for cyber attacks and weapons, against prompt injection, and against the model acting beyond what its user intended. Trusted defenders and Google’s own teams get Argon without its cyber guardrails, so they can use its full ability to find and patch software flaws.

What does Gemini 4 Argon mean for UK marketing teams?

Gemini 4 Argon changes nothing for UK marketing teams this month, because almost none can use it yet. When access opens, its strongest results match this kind of marketing work.

Analysis and reporting. Argon’s leads on Google’s finance and business-automation tests point at the work behind a good monthly report: turning exported campaign data into a clear account of what changed. Its long-answer limit also allows a full content audit in one pass, as long as you cap it.

Factual content. Argon appears more willing to say it does not know, which would make it a safer starting point for product copy, comparison pages and anything a customer or regulator could check. Someone still needs to verify the facts.

Video. Google says Argon can pick out details from long videos, and its own LVBench result supports that. Reviewing webinar recordings or a competitor’s video ads for messaging is a realistic first test.

Agents that read the web. If you plan to let an AI agent read web pages or inboxes for you, prompt injection matters: a hidden instruction planted in a page, email or document that tries to hijack the agent. In Google’s own chart for Gray Swan’s indirect prompt injection test, attacks succeeded 0.7% of the time against Argon, against 1.0% for Claude Opus 5.5 and Claude Fable 5.1 and 8.5% for GPT-6 Astra, as VentureBeat reported. Low is not zero, so limit what any agent is allowed to do.

Search is one to watch rather than act on. Google has not said Argon will power AI Mode or AI Overviews; on 3 October 2026, Gemini 3.8 Flash was the newest model in AI Mode, for Google AI Pro and Ultra subscribers. What gets a brand cited in AI answers, clear and well-sourced answers to real questions, does not depend on which model Google runs, and that is where our answer engine optimisation work starts.

The useful thing to do now is prepare a fair test:

1. Pick one recurring job. A monthly Google Ads or Meta performance commentary works well, because the inputs are exported data and the output is easy to judge.
2. Write the checklist first. List what an approved version must contain, so the same standard applies to every model.
3. Run it on your current model now. Record the model cost, the review time and how many runs failed.
4. Run the same job on Argon when it opens. Cap the output length and watch how much text it produces.
5. Compare cost per approved deliverable. Count model charges plus review time and rework, not the price per token.

Should you wait for Gemini 4 Argon?

No. Do not pause anything for it. Gemini 4 Argon has no public release date, and independent tests put it level with GPT-6 Astra and behind Claude Opus 5.5, not ahead of them.

If you use Claude Opus 5.5 or GPT-6 Astra for analysis, legal or finance-style work, plan a side-by-side test for when the API opens: that is where Google’s results are strongest, and its introductory price undercuts both until the introductory period ends. If coding is your main use, the evidence does not yet justify switching. If you pay for Google AI Ultra, watch for Argon there first.

If you want a fair test of Gemini 4 Argon against the models you already use, talk to AIWIZ. Bring one process, and we will help you measure approved work before you commit to a new model. Our AI adoption services can then build the winner into your workflows.