How to Use Gemini 4: The Beginner's Guide (First Prompt, Thinking Levels, and the Pro/Ultra Line)

CodexFuse Team · 2026-10-29T00:34:08.015Z · 15 min read

Learn how to use Gemini 4 (Argon): check Pro/Ultra access, pick the model in the app, choose a thinking level, and use it in AI Studio, Antigravity, and Copilot.

Learning how to use Gemini 4 is mostly about learning one new check: before you pick a model, prompts, or settings, you have to confirm you are on a plan that actually includes it. Gemini 4 Argon is Google DeepMind's newest frontier model, and it is the first model in the Gemini 4 family. It is built for long, multi-step work such as coding, finance, legal analysis, cybersecurity, and science, rather than quick chat replies. This beginner guide walks you through the whole first-use system in order: checking your access, picking Argon in the Gemini app, writing a first prompt that works, choosing a thinking level, using the API and AI Studio branch, and understanding the Pro versus Ultra line and the API price change on December 1, 2026.

If you have used the Gemini app before, most of the interface will feel familiar. The model picker, the prompt box, and the conversation history work the way they always have. What is new is the access gate and the thinking control, so this guide spends its time there. Our Gemini 4 Argon explainer covers what the model is and what it is built for; this guide is the hands-on companion that gets you from sign-in to a finished first task.

What is Gemini 4, and how do you use it?

Gemini 4 is Google's newest generation of frontier AI models, and Gemini 4 Argon is its first model, so for now "Gemini 4" and "Gemini 4 Argon" point to the same thing. You use it by signing in with an eligible paid plan, choosing Argon from the model picker in the Gemini app, in AI Mode in Search, in Google AI Studio, or in an agentic coding tool, and then giving it one clear task at a time. Argon accepts text, images, video, audio, and PDF input and returns text, with a shared context window of 1,048,576 tokens for input, thinking, and output.

1. Check your plan: confirm you have Google AI Pro or Ultra, a paid Gemini API tier, or another eligible route from the access table below.
2. Open the Gemini app, update it if it offers an update, and open the model picker to look for Argon.
3. Pick Argon, write one focused first prompt with the task, the context, and the format you want, and send it.
4. Read the answer critically, ask a follow-up that fixes what is wrong, and only then move on to a second task.
5. When a task gets long or hard, raise the thinking level instead of rewriting the prompt to be longer.

Do I have access to Gemini 4, or do I need to upgrade?

Check this first, because it is the most common reason a beginner cannot find Argon at all. Argon is rolling out in the Gemini app for Google AI Pro and Ultra subscribers, and it is available to paid Gemini API developers, Gemini for business customers, and eligible GitHub Copilot plans. Free Gemini accounts, Google AI Plus, and the free API tier do not include it, and Google has not announced a date for free access.

If the model picker does not show Argon, do not reinstall the app three times hoping for a different result. First confirm your plan in your Google account, then update the app, then check again, since the rollout is gradual. If you are on the free plan or Google AI Plus, the fix is a subscription change, not a settings change, because no setting unlocks a model your plan does not include.

How do I use Gemini 4 in the Gemini app?

Open the Gemini app on the web or your phone, sign in with your Google AI Pro or Ultra account, open the model picker, and choose Argon. This is the simplest version of how to use Gemini 4 Argon: one app, one picker, one prompt. Then treat your first prompt like a brief to a capable colleague: one task, enough context to do it, and the exact format you want back. A vague prompt such as "help me with my report" forces the model to guess, while a specific prompt lets Argon's longer reasoning actually help.

Try one of these as your first prompt, adapted to your own material. Each one names the task, the audience, and the shape of the answer, which is the whole prompt formula this guide recommends.

• "Summarize this contract clause in plain English for a non-lawyer, then list the three obligations it creates for me in a short table."
• "Explain the difference between supervised and unsupervised learning with one practical example of each, a comparison table, and five questions I can use to test my understanding."
• "Review this code for bugs and security problems, explain each issue in one sentence, and give me a corrected version I can test."

After the first answer, do the step beginners skip: verify before you reuse. Check any dates, numbers, names, and quoted rules against your own source, and ask a follow-up such as "Show me which part of my document supports that claim" when an answer matters. Argon is built for deep work, but deep work still needs a human check at the end.

What are Gemini 4 thinking levels, and which one should I pick?

Gemini 4 thinking levels control how long the model reasons before it answers, with three settings: low, medium, and high, and medium as the default. The level changes how much reasoning goes into the answer, not how smart the model is, so high is not "better" in general — it is slower and, on the API, more expensive, because thinking tokens bill as output tokens.

The practical rule is simple: start on medium for everything, drop to low when you are iterating quickly on drafts, and raise to high only when medium gives you a shallow or wrong answer on a genuinely hard task. If you find yourself raising the level for every prompt, the problem is usually the prompt, not the level — split the task into smaller pieces first.

How do I use Gemini 4 in Google AI Studio and the API?

Developers use Gemini 4 through Google AI Studio or the Gemini API on the paid tier, with the model name gemini-4-argon. In AI Studio, sign in with a Google account that has API billing turned on, open the model selection, and choose Gemini 4 Argon before you write your prompt. In code, set the model to gemini-4-argon and send your prompt the same way you would for any Gemini model, then read the text response back.

Moving existing code from an earlier Gemini model needs a few changes that Google's developer guide calls out. Pick a thinking level rather than tuning temperature, top_p, or top_k, because Argon ignores those settings. Remove any explicit caching calls, since Argon supports automatic caching only. Raise small output limits, because if your maximum output is under 32,768 tokens, Argon can spend more of it on thinking than you left room for. Finally, plan for long runs: Google recommends HTTP timeouts of at least 15 minutes and background execution for outputs over 64,000 tokens, since the default output cap is 65,536 tokens and longer work runs as a background job.

Keep your API key out of prompts, shared documents, screenshots, and code you publish. Store it as an environment variable or in your platform's secret store, and never paste a real key into an example you share with someone else. Our Google AI Studio beginner guide covers the key setup and first prompt in the Studio itself, if you want the click-by-click version of this branch.

How do I use Gemini 4 in Antigravity and GitHub Copilot?

Gemini 4 in Antigravity is the agentic coding route: Ultra subscribers and eligible Gemini Enterprise customers sign in to Antigravity and pick Argon for long, multi-step coding work such as large refactors and codebase migrations. Google says its own teams raised agentic code submissions by more than 35 percent while cutting code rollbacks by nearly a third when using Argon in Antigravity. Treat those as Google's claims about its own teams, not a promise about your project, and review every change an agent makes before you merge it.

In GitHub Copilot, eligible users on Pro+, Max, Business, and Enterprise plans pick Gemini 4 Argon from the model picker in their editor. That route is useful when you already live in VS Code or another supported editor and want Argon as one model among your coding models, billed on a usage basis at Google's list price. If you are choosing between the two routes, the rule is straightforward: Antigravity is for agent-driven, multi-step coding sessions, while Copilot is for Argon inside the editor you already use every day.

How much does Gemini 4 cost?

Gemini 4 pricing has two separate tracks, and confusing them is the second most common beginner mistake after the access gate. In the Gemini app, you pay a monthly subscription: Google AI Pro costs $19.99 per month and Ultra starts at $99.99 per month, and both include Argon in the app and in AI Mode in Search. On the API, you pay per token instead, and an app subscription does not cover API use — API billing is separate and must be turned on.

The December 1 change matters for budgeting, because every API rate doubles. Price any job that will run past November at the December rates, not the launch rates. Cached input gets a 95 percent discount at both prices, and prompts of at least 4,096 tokens are cached automatically, which is why repeating the same long system prompt across many calls costs far less than it first appears.

Here is a worked example you can check with a calculator. Suppose your app sends 500 requests, each with 4,000 input tokens and 1,500 output tokens. Through November 30, the input totals 2,000,000 tokens at $2 per million, which is $4, and the output totals 750,000 tokens at $10 per million, which is $7.50, for a total of $11.50. From December 1, the same job costs $8 for input plus $15 for output, or $23 in total. The lesson is not the exact total — your token counts will differ — it is that thinking-heavy jobs grow the output side fastest, so high thinking levels deserve a budget check before a big run.

Gemini 4 vs Gemini 3: which model should I pick?

Pick Argon when the task is long, hard, or high-stakes, and pick a Gemini 3 model when the task is quick, routine, or speed-sensitive. In Google's current lineup, Argon sits above Gemini 3.1 Pro and Gemini 3.8 Flash, which makes it the frontier choice for deep work rather than the everyday default. Our Google Gemini beginner guide covers the app basics that apply whichever model you choose, and the model picker is where you switch between them.

One honest comparison note: Argon's rival in this generation is GPT-6 Astra, Google's competitor's flagship, and the two trade benchmark wins depending on the test. If you already pay for one ecosystem, that matters more than any single benchmark table, because switching ecosystems costs you your history, integrations, and habits. Our GPT-6 Astra explainer covers the rival model if you want the comparison before you commit.

What should I check before I trust a Gemini 4 answer?

Argon answers more directly and hedges less than earlier Gemini models, according to Google, which makes verification more important, not less. A confident answer is not a verified answer, so check dates, numbers, names, quoted rules, and code by running it or comparing it with your source document. Argon's knowledge cutoff is March 2026, so anything that changed after that date — prices, plans, laws, and product details — needs a fresh source, not the model's memory.

For long tasks, give the model the full source material instead of a summary of it, since the 1-million-token window exists for exactly that job. Then break the work into checkpoints: ask for an outline first, check the outline, then ask for each section, and check each one before moving on. Finally, keep sensitive material out of prompts you do not need to share — no passwords, API keys, private customer data, or unreleased company documents — because a prompt sent to a hosted model is data you have handed to someone else's system.

Sources

Lindy, "Gemini 4 Argon: Who Can Use It Now, Pricing, and Benchmarks" (lindy.ai) — announced September 30, 2026, generally available October 13, 2026; first Gemini 4 model; 1,048,576-token shared context window; 65,536-token default output cap with longer work as a background job; text, image, video, audio, and PDF input with text output; thinking levels low, medium (default), and high; knowledge cutoff March 2026; API pricing $2 input and $10 output per million tokens through November 30, 2026, then $4 and $20 from December 1, 2026, with cached input at $0.10 rising to $0.20; access table for the Gemini app, AI Mode, Antigravity, the API and AI Studio, Gemini for business, GitHub Copilot, and the Fairwind Program; Pro $19.99 per month and Ultra $99.99 or $199.99 per month; API migration notes including ignored temperature, top_p, and top_k settings, automatic caching only, raising output limits under 32,768 tokens, and 15-minute HTTP timeouts for long runs; Google's Antigravity claims of more than 35 percent higher agentic code submissions and nearly a third fewer rollbacks. SeoDigix, "Gemini 4 Argon: Features, How to Use & Pricing (2026)" (seodigix.com, updated October 20, 2026) — entry points by user type including the Gemini app and AI Mode for Pro and Ultra, Antigravity for coders, and AI Studio and the Gemini API on the paid tier with model name gemini-4-argon; Pro versus Ultra comparison including 5 TB versus 20 TB of storage and 5x usage limits on Ultra; API pricing table matching the November 30 and December 1 rates; 10,000 scientists offered Antigravity's highest usage tier free for six months subject to selection. Google Cloud, "Get started with Gemini 4" (docs.cloud.google.com) — the first Gemini 4 model is Gemini 4 Argon, built for deep work in coding, finance, legal, cybersecurity, and science; developer quickstart with model name gemini-4-argon and availability on the global endpoint and in the us and eu multi-regions.

Frequently asked questions

Is Gemini 4 free?

No. In the Gemini app, Argon requires Google AI Pro or Ultra, and on the API it requires the paid tier. Free Gemini accounts, Google AI Plus, and the free API tier do not include it, and Google has not announced a date for free access.

Why can't I see Gemini 4 Argon in my app?

Usually because your plan does not include it, or because the rollout has not reached your account yet. Confirm you have Google AI Pro or Ultra, update the app, and check the model picker again. If you are on the free plan or Google AI Plus, upgrading the plan is the fix, not a settings change.

Which Gemini 4 thinking level should I use?

Start on medium, which is the default and fits most work. Drop to low for quick drafts and fast checks, and raise to high only for hard mathematics, difficult coding, or long research, since high thinking takes longer and costs more on the API because thinking tokens bill as output.

Does Google AI Pro include Gemini 4 API access?

No. Pro includes Argon in the Gemini app and in AI Mode in Search, but API use is billed separately per token and needs API billing turned on. Keep the two tracks separate when you budget: a monthly subscription for the app, and per-token billing for the API.

Will Gemini 4 come to the free plan?

Google has not said so. Argon launched as a paid-plan model, and the honest plan is to assume it stays paid until Google announces otherwise. If you want to try it without committing, a one-month Pro subscription tested on your real work is a fairer trial than any benchmark table.