Gemini 4 Argon is Google’s newest frontier AI model, opened broadly on October 13, 2026 and built for long, complex work in coding, finance, legal analysis, cybersecurity and science. It reads text, images, audio and video, but it answers in text only. It is not free: you need a paid Google AI plan or a paid API account.
Early coverage has aged badly. Articles written while only cybersecurity defenders had access said it was gated, and many claimed Argon writes a million tokens in one reply. Neither is accurate today, and a reader who trusts those pages will pick the wrong plan or budget the wrong amount.
This guide is based on Google’s published documents, checked on October 16, 2026, not on hands-on testing. It sets out what Argon is, what it costs, where sources disagree and which reader types should act on it now.
What is Gemini 4 Argon?
Gemini 4 Argon is a frontier model from Google DeepMind, released broadly on October 13, 2026. Google says it excels at deep work on complex, real-world tasks across coding, finance, legal, cybersecurity and science. It accepts text, images, audio and video as input and produces text as output, with a knowledge cutoff of March 2026.
Google’s Gemini 4 Argon model card adds two details that matter. Argon “is not a modification or a fine-tune of a prior model,” so do not treat it as Gemini 3 with a new label. The card also lists limitations: hallucinations, occasional slowness or timeouts, and knowledge that is less current in some domains.
The release came in stages. Google first limited Argon to vetted cyber defenders in its Fairwind Program, then opened it to paid users on October 13. Cybersecurity teams can still apply for a version without cyber guardrails, according to Google’s Gemini 4 Argon launch post. If you want the wider product context first, our guide to what Gemini is covers the family before Argon.
What are the key specs?
This table collects the numbers readers ask for most, taken from Google’s model card, long decode documentation and pricing page.
| Spec | What Google publishes |
|---|---|
| Release | October 13, 2026 (broad availability) |
| Knowledge cutoff | March 2026 |
| Input | Text, images, audio, video |
| Output | Text only |
| Context window | 1,048,576 tokens, shared by input, thinking and output |
| Output per request | 65,536 tokens by default; more only through long decode in background mode |
| Thinking | Adjustable thinking level; thinking tokens are billed as output |
| Free API tier | None |
What does Gemini 4 Argon do differently from earlier Gemini models?
Google claims three changes: it handles longer, multi-step tasks, it follows style instructions more closely, and it hedges and repeats itself less. Those are Google’s own descriptions, not independent findings. The model card lists the intended uses it was built for, and they are narrower and more professional than a general chatbot.
The intended uses on the model card are:
- Deep, multi-turn investigations
- Complex engineering
- Deep visual understanding
- Multimodal document analysis and synthesis
- Legal and financial analysis
- Scientific and mathematical reasoning
- Cybersecurity defense
- Writing
The practical meaning: Argon is aimed at jobs where a person would otherwise spend hours reading, cross-checking and drafting. It is not positioned as a faster way to answer a quick question. Google’s launch post says responses are more direct and hedge less.
What does Gemini 4 Argon cost?
Through November 30, 2026, Standard API pricing is $2.00 per million input tokens and $10.00 per million output tokens. From December 1, 2026 it rises to $4.00 and $20.00. There is no free API tier, and the Gemini app version requires a paid Google AI plan.
Google’s Gemini API pricing page lists four paid tiers. Output prices include thinking tokens, so a model that reasons at length costs more than its visible answer suggests.
| Tier | Input, through Nov 30 | Output, through Nov 30 | Input, from Dec 1 | Output, from Dec 1 |
|---|---|---|---|---|
| Standard | $2.00 | $10.00 | $4.00 | $20.00 |
| Batch | $1.00 | $5.00 | $2.00 | $10.00 |
| Flex | $1.00 | $5.00 | $2.00 | $10.00 |
| Priority | $3.60 | $18.00 | $7.20 | $36.00 |
Prices are USD per million tokens, as of October 16, 2026. To see what the change means in practice, take a team that sends 10 million input tokens and receives 2 million output tokens a month on the Standard tier. That costs $40 at the introductory rate (10 x $2 plus 2 x $10) and $80 from December 1. The usage did not change; the bill doubled.
For app users, Google’s Gemini plans page shows Google AI Pro at $19.99 a month and Google AI Ultra from $99.99 a month. Here the sources disagree. The launch post names only Pro and Ultra for the Gemini app, yet the plans page also lists Google AI Plus ($4.99 a month) as including Argon. Trust the launch post for what is rolling out, and check inside the app before paying for Plus. Our Gemini 4 Argon pricing guide breaks the numbers down further, and the older Gemini pricing overview covers the rest of the lineup.
Budget At The December Price
Model every pilot at $4.00 input and $20.00 output, not the introductory rates. If the business case only works at $2.00 and $10.00, it expires on November 30, 2026. Batch and Flex are half the Standard price, so route anything that can wait through them.
Who can use Gemini 4 Argon today?
Anyone with a paid route. Google AI Pro and Ultra subscribers get Argon in the Gemini app and in AI Mode in Search. Ultra adds Antigravity and AI Studio. Developers and enterprises use the paid Gemini API. The free Gemini tier does not list Argon, and neither does the free API.
The table shows each route and its catch. Our guide to accessing Gemini 4 Argon walks through the steps for each one.
| Route | Who gets it | The catch |
|---|---|---|
| Gemini app | Google AI Pro and Ultra | Rolling out from October 13; usage limits apply |
| AI Mode in Search | Google AI Pro and Ultra | Tied to the same plans |
| Antigravity | Google AI Ultra (individuals) and enterprise plans | Not an entry-plan feature |
| AI Studio | Ultra subscribers and paid API accounts | Google’s pricing page says Argon is not on the Pro plan or the free tier |
| Gemini API | Paid tier | No free tier; token costs rise on December 1 |
| Fairwind Program | Vetted cybersecurity defenders | Application and approval required |
| Scientists program | Up to 10,000 scientists | Free Antigravity top tier for six months, subject to eligibility, capacity and selection |
How does Gemini 4 Argon compare with GPT-6 Astra and Claude Opus 5.5?
On Google’s own model card, Argon leads 13 of 19 benchmark rows, ties one and trails five against GPT-6 Astra, Claude Fable 5.1 and Claude Opus 5.5. Its wins cluster in finance, legal, long-context and visual tests. Its losses cluster in terminal and some software engineering tests. Google published these results itself.
Some examples from that table show the pattern. Argon scores 19.6% on the Harvey Legal Agent Benchmark, against 6.7% for the next best model. It scores 65.4% on Vals Finance Agent v2 against 58.9%, and 84.2% on GraphWalks at 256k to 1M tokens against 71.8%. It trails on Terminal-bench 4.0 (57.4% against 66.4% for Opus 5.5), FrontierSWE v2 (55.0% against 65.5% for Astra) and Terminal-Bench Science 0.1 (57.6% against 68.1% for Astra).
Treat the card with care. It does not state the thinking settings used, and it adds no footnotes on how competitor scores were obtained. Independent leaderboards do not all rank Argon where Google does, so wait for a second source before you stake a purchase on a single row. We cover the numbers in our Gemini 4 Argon benchmarks guide and the head-to-head in our Argon vs GPT-6 Astra vs Claude Opus 5.5 comparison. For the rivals on their own terms, see GPT-6 Astra and Claude Opus 5.5.
What is Gemini 4 Argon not?
Argon is not free, not a one-million-token reply machine and not an image or audio generator. It is also not the whole Gemini 4 family, and its benchmark lead is not independently confirmed. This table corrects the most common misreadings and says which source to trust when pages disagree.
| Common misreading | What the documents say | Source to trust |
|---|---|---|
| “Argon is free” | No free API tier. Free Gemini app users do not see it listed. Paid plans or paid API accounts only. | Google’s pricing and plans pages |
| “It writes 1M tokens in one reply” | The default is 65,536 output tokens per request. Longer output needs long decode in background mode, which runs the model in 64K hops, shares the 1,048,576-token window with your input and bills every hop. | Google’s long decode documentation, not headlines |
| “It makes images, audio or video” | It reads all of them but outputs text only. | Model card |
| “Only cyber defenders can use it” | True until October 13, 2026. Now available to paid users and developers; Fairwind remains for the guardrail-free version. | The dated Google launch post, over articles from before October 13 |
| “It is the whole Gemini 4 family” | Google calls Argon “just the beginning” and refers to “the rest of the Gemini 4 family,” so more models are coming. | Google launch post |
| “The benchmarks prove it is the best” | The scores are Google-reported, show no thinking settings and include five rows where Argon trails. | Read the card, then wait for independent results |
| “It knows this year’s news” | Knowledge cutoff is March 2026. | Model card |
One caution on the “first model” row. Google never writes “first release” in the pages we read. It says more Gemini 4 models are coming, so reading Argon as the opening member of the family is a fair inference, not a stated fact.
Who should care about Gemini 4 Argon, and who can ignore it?
Care now if you analyze long documents, build agents or work in legal, finance or security. Watch if you mostly write, summarize or chat. Ignore it for now if you only use the free Gemini app, need generated images or audio, or cannot pay for token-heavy work. Most learning leaders land in the watch group.
The table below sorts reader types by what Argon offers them. The verdicts are our judgment from Google’s documentation, not Google claims.
| Reader | Verdict | Why | First step |
|---|---|---|---|
| L&D leader choosing tools | Watch | Argon is a model, not an LMS feature. Its value reaches you only when your vendors adopt it. | Ask each vendor which model powers its AI features |
| Instructional designer with long source material | Care, pilot it | Multimodal document analysis is an intended use. Output is text, so course media comes from elsewhere. | Run one policy or SME document set through the app |
| Compliance or legal training owner | Care, with expert review | Legal analysis is an intended use, but hallucinations are a listed limitation. | Use it for first-pass comparison and keep SME sign-off |
| Developer or learning-tech engineer | Care now | Paid API, large context and long decode suit agents and large rewrites. | Prototype and cost it at the December price |
| Security team | Care now | Fairwind gives defenders a version without cyber guardrails. | Apply through Google’s program |
| Researcher or scientist | Care, apply | Google offers 10,000 scientists a free six-month Antigravity tier. | Check eligibility before capacity fills |
| Casual user on the free Gemini app | Ignore for now | The Free tier does not list Argon. | Revisit if Google changes the plan lineup |
| Team that needs image, audio or video output | Ignore | Output is text only. | Choose a model that generates media |
| Team running short, high-volume tasks | Check first | Google positions Argon for deep work, and reasoning tokens add cost. | Test whether a lighter model is enough |
If your reader type is unclear, use three questions. Does the task involve long documents or multi-step work? Can you pay for an access route? Is text output enough? Three yeses mean a pilot is worth the time. Any no means you can wait for the rest of the Gemini 4 family and for independent results.
How should a learning team use Gemini 4 Argon without over-trusting it?
Use it for drafting and first-pass analysis, and keep a subject matter expert in the loop for anything that carries risk. Google’s own model card lists hallucinations, slowness and timeouts as limitations, so output is a draft until a person checks it against the source.
Three uses map to the model card’s intended uses. These are our suggestions for learning teams, not Google examples:
- Turning a long policy PDF or recorded walkthrough into draft learning objectives for an instructional designer to edit
- Comparing two versions of a procedure and listing what changed, ahead of a training update
- Producing a first-pass summary of source material before an SME interview
Keep compliance and safety content out of unattended workflows. Where a regulation applies, the training owner should check the regulator’s text directly, because a fluent answer is not a verified one. Our guide to which AI model fits which learning job applies the same test to two rival models.
Ask For Gaps Not Just Answers
Because Google says Argon hedges less, a confident tone tells you nothing about accuracy. Finish each task with a second prompt asking it to list every claim it could not support from the supplied documents. That list is your SME review queue.
What should you do next?
Pick the one route that matches your reader type in the table above and try a single, real task before you commit budget.
Write your prompts before you open the app. Long, multi-document tasks reward clear instructions, and our guide to prompting Gemini 4 Argon shows how to structure them.
Finally, set a calendar reminder for November 30, 2026. If you piloted at the introductory price, rerun your cost estimate at the December rates before you roll anything out.
FAQ
Q1. What is Gemini 4 Argon?
Gemini 4 Argon is a Google DeepMind frontier model released broadly on October 13, 2026. Google built it for deep, long-running work in coding, finance, legal analysis, cybersecurity and science. It accepts text, images, audio and video as input but produces text only, and its knowledge cutoff is March 2026.
Q2. Is Gemini 4 Argon free?
No. Google’s pricing page lists no free API tier for Argon, and the Gemini plans page does not list it for the Free plan. You need a paid Google AI plan for the Gemini app or a paid API account. Google offers a free six-month Antigravity tier to 10,000 scientists, subject to selection.
Q3. When was Gemini 4 Argon released?
Google opened Gemini 4 Argon broadly on October 13, 2026. Before that, access was limited to vetted cybersecurity defenders in the Fairwind Program. Articles written before October 13 that call Argon gated or defenders-only describe the earlier stage, not the current one.
Q4. What is the Gemini 4 Argon context window?
The context window is 1,048,576 tokens, shared by input, thinking and output. Each request returns up to 65,536 output tokens by default. Longer output needs long decode in background mode, and the extra tokens come out of the same shared window, so a very long prompt leaves less room for the answer.
Q5. Can Gemini 4 Argon generate images or audio?
No. Argon can read text, images, audio and video, but its output is text only, according to Google’s model card. If your work needs generated images, audio or video, you will need a different model for that step and can use Argon for the analysis and writing around it.
Q6. Is Gemini 4 Argon better than GPT-6 Astra?
It depends on the task. Google’s model card shows Argon leading 13 of 19 benchmark rows, mostly in finance, legal and long-context tests, and trailing GPT-6 Astra on rows such as FrontierSWE v2. Google published those results itself, so treat them as vendor-reported until independent testing confirms them.
Q7. Is Gemini 4 Argon the only Gemini 4 model?
Not for long. Google’s launch post calls Argon “just the beginning” and refers to “the rest of the Gemini 4 family,” but it names no other models. Argon is the one you can use today, so plan around its documented limits instead of waiting for undisclosed siblings.
