Category: Uncategorized

You have a client meeting on Thursday. Everything you need is somewhere in one folder: call notes, an old quote, three PDFs, a spreadsheet and a long email thread. Turning that pile into a clean proposal deck usually eats an evening. Gemini 4 Argon, Google’s most capable model you can use today, can do the first draft of that job for you, and you get your evening back.

In short: Gemini 4 Argon is Google’s most capable model you can use today, built for long, multi-step work across documents, spreadsheets and software projects. You get it in the Gemini app with a Google AI Pro or Ultra plan, and developers can pay per use through the Gemini API. Below: what changed, five uses by role, a 10-minute test with a ready-to-copy prompt, and the honest limits. One heads-up: on 28 October Google announced Gemini 4 Carbon, an even stronger model, but for now only a small group of security partners can use it. Argon is the one you can open today.

What changed with Gemini 4 Argon

On 13 October 2026 Google made Gemini 4 Argon available worldwide. It sits in the Gemini app and in AI Mode in Search for Google AI Pro and Ultra subscribers, in Google Antigravity, Google’s agent platform, for Ultra subscribers, and in the Gemini API for developers on the paid tier. Here is what that means at your desk:

• It handles messy input. In Google’s own demo, Argon takes an unstructured folder of project material and formats it into a clean proposal deck for a client. For you: less copy and paste between files before you can even start.
• It follows your style instructions better. Google says Argon follows style instructions better than any earlier Gemini model, answers more directly with less hedging, and skips clichés. For you: fewer rewrites to make it sound like your company.
• Your plan already reads a lot at once. This part is not new with Argon: Pro and Ultra already gave the Gemini app a context window of 1 million tokens, which is the amount of material it can keep in mind together. What Argon adds is better work on top of all that material. For you: a full contract, a year of meeting notes or a big spreadsheet in one chat, now handled by a stronger model.

How your team can use it

Sales: a proposal deck from a client folder

Your account manager has a folder for a prospect: discovery call notes, pricing from last year and the brief they sent. She uploads it, asks Argon for a proposal deck in the team’s usual order (problem, approach, timeline, price, next step) and then asks for a short follow-up email to send with it. She checks the numbers and sends it herself.

Finance: long documents and spreadsheets side by side

A controller has the quarterly budget sheet and a 40-page board pack. He asks Argon to list every place where the spending in the sheet does not match the story in the pack, with the page and the cell.

Operations and managers: a clean handoff

A project lead is going on leave. She gives Argon the project folder and asks for a handoff document: open tasks, who owns what, decisions made, risks and the three things her colleague must do in week one.

HR: check a new policy against the old handbook

An HR advisor uploads the current employee handbook and the draft of a new hybrid work policy. She asks Argon to list every sentence in the draft that conflicts with the handbook, quoting both. She still decides what changes, without the highlighter marathon.

Support: find the pattern in a ticket export

A support lead exports last month’s tickets as a spreadsheet. He asks Argon to group them by root cause, count each group and write a one-page summary for the product team. The counting is the part you double-check before you share it.

Before vs now

• Last month: an evening opening ten files and building slides from scratch. Now: upload the folder, give your structure, edit a draft.
• Last month: a first draft from a long report that still needed heavy rewriting to match your style. Now: the same long report and spreadsheet in one chat (the 1 million token room comes with Pro and Ultra, as before), with Argon following your style instructions more closely.

Try it yourself in 10 minutes

Pick one real client or project you are working on this week. Then:

1. Open the Gemini app or the Gemini website and sign in with the Google account that has your AI Pro or Ultra plan.
2. Open the model picker at the top and choose 4 Argon. Don’t see it? Update the app and check again later, since Google rolls features out in stages.
3. Upload the files for that one client: notes, the brief, an old quote, the relevant spreadsheet. Leave out confidential files you don’t need.
4. Paste the prompt below and fill in the brackets.
5. Read the outline first. Fix the order or the tone in a reply before you ask for the full deck.
6. Turn the outline into real slides. On the Gemini website, click Canvas below the chat box and type: “Create a presentation from the approved outline above.” Gemini builds the slides with a theme and images in the Canvas panel. Click Export at the top right to open them in Google Slides, where you can edit and share the deck.
7. Check every number, date and name against your files before anything goes out.

Copy this prompt:

You are helping me prepare a client proposal. The files I uploaded are everything I have on [client name]: call notes, their brief and older material.

1. First, list in 5 bullets what the client actually wants, using only what is in the files. Mark anything you are unsure about with “CHECK”.

2. Then make an outline for a proposal deck of at most 10 slides in this order: their situation, the problem, our approach, timeline, investment, next step.

3. Write it in plain, friendly English, short sentences, no buzzwords. Use our prices only if they appear in the files; otherwise write [PRICE].

4. Finish with a 120-word follow-up email I can send with the deck.

Wait for my feedback on the outline before writing full slides.

The “CHECK” and [PRICE] markers matter. They make the model show you where it is guessing, so nothing invented slips into a client document. If you want the slide design itself to look good, our guide on how to use AI to make a presentation covers that part.

Honest limits

• It is not free. In the Gemini app you need Google AI Pro or Ultra. Google AI Plus does not list Argon, and The Verge noted it is not clear when, or if, Plus and free users will get it.
• Usage has a ceiling. Gemini app limits are based on how heavy your requests are. They refresh every 5 hours until you reach a weekly limit. Pro gets 4 times the standard limits; Ultra gets 5 or 20 times the Pro limits, depending on the plan.
• Busy moments. Some Pro users on Google’s help forum report a message that a different model answered because advanced models were in high demand.
• Not every country. Argon in AI Mode is limited to select countries, and some features, like Tasks, start in select countries only.
• You still check the facts. A model that writes clean slides can still misread a number or mix up two clients. Review everything before it goes out under your name.
• API costs will double. The developer price is introductory until 30 November 2026, then it goes up.

Who can use Gemini 4 Argon now

Checked on 1 November 2026 on Google’s own plan and pricing pages (US prices; in the Netherlands, for example, Pro shows as €21.99 a month).

Plan	Price	What it allows for work
Free	$0	No Argon access
Google AI Plus	$4.99 a month	Argon not listed in the plan
Google AI Pro	$19.99 a month	Argon in the Gemini app and AI Mode, 1M token context, 4x standard usage limits
Google AI Ultra	$99.99 or $199.99 a month	Highest Argon access (5x or 20x Pro limits), plus Argon in Google Antigravity
Gemini API (developers)	$2 per 1M input tokens and $10 per 1M output tokens until 30 Nov 2026; $4 and $20 from 1 Dec 2026	Build Argon into your own tools; paid tier only, not on the free tier

Quick questions

Is Gemini 4 Argon free?

No. In the Gemini app it comes with Google AI Pro or Ultra, and in the Gemini API it is only on the paid tier.

Do I need Ultra, or is Pro enough?

Pro is enough for the Gemini app and AI Mode. Ultra gives higher limits and adds Argon in Google Antigravity.

How much does Gemini 4 Argon cost in the API?

$2 per million input tokens and $10 per million output tokens until 30 November 2026. From 1 December it is $4 and $20, and thinking tokens count as output.

Can Argon send the proposal to my client by itself?

Google’s demo shows Gemini drafting and emailing a deck and setting up a scheduled task. Whether you see those options depends on your plan and country, and it is wise to send client emails yourself after a check.

New to Gemini altogether? Start with our Gemini tutorial for beginners. Working mostly with numbers? See how to analyze a spreadsheet for work with AI. And for more practical AI guides you can use this week, start with the free Launchpad on our home page.

Sources

• Google: Gemini 4 Argon, built to solve your hardest problems (13 Oct 2026)
• Google: Gemini 4 Argon for Robotics, with getting-started details (13 Oct 2026)
• Google: Gemini Developer API pricing (checked 1 Nov 2026)
• Google: Google AI Pro and Ultra plans (checked 1 Nov 2026)
• Gemini Apps Help: limits and feature availability
• Google: Introducing Gemini 4 Carbon (28 Oct 2026)
• Google Workspace Updates: Generate presentations in the Gemini app
• The Verge: Google finally releases Gemini 4 Argon (13 Oct 2026)
• 9to5Google: Gemini app rolling out Gemini 4 Argon to AI Pro and AI Ultra (13 Oct 2026)
• Gemini Apps Community: Pro user report on model fallback

That supplier quote sitting in your drafts. The dentist booking you keep meaning to make. The tool signup that wants a confirmation email. Each one eats a slice of your afternoon, and half the replies land in the wrong folder.

Imagine handing those loops to your assistant with one sentence. With a Grok Bot email, your Bot claims its own inbox, writes out, and the replies come back to it instead of you.

In short: On 9 October 2026, Grok Bot got its own email at @mail.grokbot.com. It can sign up for services, contact businesses for you, or schedule time with someone. You claim it by asking your Bot (or tagging @bot on X). Team plans need an admin to turn it on first. A 10-minute claim and handoff is below.

What changed

On 9 October 2026 the official @bot account on X announced that Grok Bot now has its own email. The Bot can use that address to sign up for services, contact businesses for you, or schedule time with someone. The rollout started the same day.

Here is what that means at your desk:

• Sign-ups without your personal inbox. When a tool asks for an email to finish setup, the Bot can use its own address and catch the confirmation itself.
• Supplier and shop emails in the Bot’s thread. Quotes, availability checks, and follow-ups can leave from the Bot’s address, so the reply does not bury your own mail.
• Booking time for you. Asking someone for a slot (a meeting, an appointment) can go out from the Bot, with the back-and-forth staying on that thread.
• One sentence to start. You ask the Bot to claim its email. Addresses end in @mail.grokbot.com, and you can request a custom name if it is free.

When you give your assistant its own email address, you can hand it the repetitive tasks, and you get time back for your family and the things you love.

How your team can use it

• Operations: “Claim your email address, then write to this supplier and ask for a quote on 200 units of SKU-14, delivery to our warehouse next month.” You review the draft or the reply instead of starting from a blank email.
• Sales: “Use your email to book a 20-minute call with this prospect next week. Offer Tuesday or Thursday afternoon.” The calendar dance leaves your inbox alone.
• Support: “Sign up for this returns portal with your email.” The signup mail goes to the Bot’s inbox, not yours.
• HR: “Email this training venue and ask for three available dates for a 12-person workshop.” You pick from the options, not from a pile of unread mail.
• Managers: “Schedule a 30-minute check-in with this contractor for Friday morning.” The Bot handles the back-and-forth until the time is set.

If you also draft everyday replies yourself, see how to use ChatGPT to draft reply emails faster. For picking the right model behind each Grok Bot job, see how to use Grok Bot to get the best AI model for every work task.

Before vs now

• Last month: every signup confirmation, supplier reply, and booking ping landed in your inbox. You chased threads between meetings.
• Now: the Bot has its own @mail.grokbot.com address. You hand it the loop in plain language. Replies come back to the Bot, and you only step in when something needs a human decision.

How to claim your Grok Bot email

Ten minutes, start to finish.

1. Open your Grok Bot chat.
2. Type this exact line: Claim your email address.
3. Pick a free name. Short and work-safe works best, like your Bot’s name. The Bot checks whether it is available, and you end up with an address like yourbot@mail.grokbot.com.
4. Tap Allow once to approve the claim.
5. On a team plan, ask your admin to enable Bot email first if the claim fails.
6. Then hand off one real loop: Using your new email, write a short note to this supplier asking for a quote on [item] by Friday. Show me the draft before you send.
7. Read the draft. Approve or edit. Let the reply land in the Bot’s inbox, not yours.

Choose the name with care: the name you pick is your Bot’s address from then on.

If you cannot open Grok Bot right now, you can also tag @bot on X and ask it to set the inbox up, per the rollout post.

Honest limits

• What it may send in your name. Outgoing mail from the Bot represents you or your team. Start with “show me the draft before you send” until you trust the loop.
• One inbox per Bot is not confirmed. The announcement talks about “its own email” and “your inbox”. SpaceXAI has not said whether every Bot you create gets a separate address, or whether the inbox follows the user. Treat it as one address for your setup until the docs say otherwise.
• Team plans need an admin. If you work inside a company seat, an admin must enable the feature before you can claim.
• No separate price announced. SpaceXAI did not list an add-on fee for Bot email. Grok Bot itself sits on paid Cursor plans from about $20 a month, or a linked SuperGrok account. Confirm the current plan page before you buy.

Who can use it now

Checked on 1 November 2026 against the @bot rollout posts and press walkthroughs:

Who	What you need	Notes
Individual Grok Bot users	Ask your Bot, or tag @bot on X	Rolling out from 9 October 2026
Team / company seats	Admin enables Bot email, then you claim	Claim fails until the admin switch is on
Address format	@mail.grokbot.com	Custom name if available (per 9to5Mac)
Extra fee for email	None announced	Grok Bot is bundled with Cursor / SuperGrok plans

Quick questions

How do I claim a Grok Bot email address?

In your Grok Bot chat, type “Claim your email address”, pick a free name, and tap Allow once. You can also tag @bot on X. On a team plan, an admin must enable the feature first.

What can Grok Bot do with its email?

The announcement names three jobs: sign up for services, contact businesses for you, and schedule time with someone. Anything beyond that is not documented yet.

Does every Bot get its own inbox?

SpaceXAI has not said. Until the docs cover it, treat the inbox as one address for your user setup, not a guaranteed address per Bot name.

Is Grok Bot email free?

No separate email fee was announced. You still need a Grok Bot seat on a paid Cursor or SuperGrok plan. Check the current plan page for the exact price.

Sources

• @bot announcement: Grok Bot now has its own email (9 October 2026)
• @bot rollout: claim by asking Bot or tagging @bot; admins must enable for teams (9 October 2026)
• 9to5Mac: how to claim, @mail.grokbot.com, custom name (9 October 2026)
• CellCog walkthrough and open questions (docs gaps, unit of the inbox) (10 October 2026)

For a wider look at desktop AI agents in the office, see best AI agent for desktop workflows in the office. And when the same chase note keeps landing back on your desk, automate follow-up emails without code so the sequence keeps moving without a developer.

You write the client email in one AI tool, make the flyer picture in another, and go looking for a third that can do a short jingle for the video. Three logins, three bills, and you are still not sure you picked the right tool for each job. Elon Musk just said Grok Bot will make that choice for you, using the best model per task.

Imagine typing one request at 9, “make the launch email and the flyer that goes with it”, and having both drafts waiting before your first coffee is cold, without ever choosing a model.

In short: Grok Bot is an AI teammate that does multi-step work for you on its own cloud computer. On 7 October 2026 Musk said it will use the best back-end model for each task, naming Claude Opus 5.5, MidJourney and Suno. It comes with paid Cursor plans from $20 a month or a linked SuperGrok account, and a 10-minute test is below.

Why this is good news for your work

• No more tool shopping. You stop asking “which AI is best for this?”. You describe the result you want, and Grok Bot picks the model for each part of the job.
• One place for mixed work. Text, images and music can come out of one request to one Bot, instead of three apps and a lot of copy and paste.
• Better results without effort. When a stronger model appears, the Bot can switch to it. You don’t have to follow every new release to get the benefit.
• Your skill shifts to what matters. Knowing exactly what you want, and checking the result, becomes the valuable part of your job. That is a skill anyone can learn.

What changed: Grok Bot picks the best model per task

On 7 October 2026 Musk posted on X. His post opens with “Important note regarding Grok Bot” and goes on: “Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs. Whatever is most likely to give you the best outcome.” Here is what that means at your desk:

• You don’t pick the model. Grok Bot’s own settings docs already say model selection is managed for you and there is no model picker. The post says the pool behind it gets wider. So the skill that matters is describing the result you want, not knowing which AI is good at what.
• More kinds of work in one request. Musk named a writing and reasoning model (Claude Opus 5.5), an image tool (MidJourney) and a music tool (Suno). If that lands as described, one Bot can write the email, make the picture and add the music, instead of you copying work between three apps.
• The mix can change. The security docs say the serving mix can change over time and no fixed set of providers is guaranteed. Results may get better without you doing anything, and they may also look a little different from one week to the next.
• You can see what was used. The same docs say usage analytics show the model that served each request, and billing follows that model.

This is a developing story. Musk’s post gives no date and no list of which task goes to which model, and when we checked on 7 October the official Grok Bot docs didn’t either.

One request, the right model for each part

This is the heart of the news. You give Grok Bot one job, and it can send each part to the model that is strongest at it. Musk named three examples. Here is how that maps onto everyday work:

Part of the job	Example request	Where it could go
Writing and reasoning	“Draft the client email and summarise these three contracts.”	A strong writing and reasoning model, such as Claude Opus 5.5
Images	“Make a clean flyer image with our headline.”	An image generator, such as MidJourney
Music and audio	“Add a calm 15-second music bed for the intro video.”	A music tool, such as Suno
Everything else	“Turn this spreadsheet into a chart and a short update.”	“Other leading APIs”, whatever fits the task best

These pairings are examples based on what each tool is known for. Grok Bot makes the choice itself, and xAI has not published an official list of which task goes to which model yet. The point for you stays the same: describe the finished result, not the tool.

How your team can use it

A Bot works by message: you give it the task, the context and access to the tools it needs, and it comes back with the work. It can stop and ask for your approval before anything goes out, but that depends on the action and your settings, so ask for drafts first. A few ways that plays out:

• Sales: “Here are my notes from yesterday’s five demo calls. Draft a short follow-up email for each one, and make one clean product image I can attach.” You read five drafts instead of writing five emails.
• HR and training: “Turn this onboarding checklist into a friendly welcome page for new starters, plus a calm 15-second music bed for the intro video.” If you also record narration for training slides, see how to use Suno Speech to make a voiceover for your training video.
• Finance: “Read these three supplier contracts and put the payment terms, notice periods and price-increase clauses in one table.” It’s reading-and-reasoning work, and you never have to decide which model is best at it.
• Support: “Rewrite this help article so a first-time customer gets it in one read, and make a simple diagram of the three steps.” Text and picture come back together, ready for you to check.
• Managers: “Turn my notes from this week into a five-bullet team update and one summary image for the Monday slide.” Same request every week, same format.

Before vs now

Last month: pick a tool for the text, another for the image, maybe a third for music. Copy the output from one to the next, fix the formatting, and pay for each one separately.

The plan now: one request to one Bot, describing the finished result. Grok Bot chooses the model behind each part. It can stop for your approval before anything goes out, depending on the action and your settings, so ask for drafts first, like the prompt below does. Whether every part already runs on the models Musk named is not confirmed yet, so treat it as a direction, not a promise for today.

Try it yourself (10 minutes)

1. Open Grok Bot on your computer or phone and start a chat with a Bot. If you don’t have one yet, create one and give it a plain job title like “Marketing helper”.
2. Copy the prompt below, fill in the parts in round brackets, and send it.
3. Read every draft. Fix one thing you don’t like by replying in plain words (“shorter, and less salesy”).
4. When the format is right, ask the Bot to use this format for the next launch too, so you don’t have to explain it again.

I run (your team) at (your company). Next week we launch (your product or event).
Make me three things:
1. A 120-word announcement email for our customers. Friendly, plain words, one clear next step.
2. A simple flyer image with the headline "(your headline)". No logos, no real people.
3. A 15-second upbeat music idea for a short promo video.
Show everything as drafts. Do not send or post anything.
At the end, tell me which of the three you could not do yet.

That last line matters. Because the new model mix is still rolling out, it tells you straight away what works today and what doesn’t.

Honest limits

• Timing is unclear. Musk said “going forward”. There is no published date, and no official list of which task goes to which model.
• You can’t force a model. There is no model picker. A team model allowlist exists only on the Enterprise plan, and the docs say enforcement is not guaranteed.
• Usage is metered. Paid plans include weekly usage that resets each week. After that, extra work can continue as on-demand usage, billed from model and token cost, if on-demand is turned on. Big image or music jobs may use more of your allowance, so start small.
• Your data goes to other providers. When a task goes to another company’s model, your text goes with it. The docs say that with Privacy Mode on, customer data is not used for training, and that providers don’t keep prompts or outputs under Cursor’s existing agreements. They also say providers may run abuse and safety checks and may store flagged data. If your contract limits which companies can see client data, ask your account team first, and keep sensitive client details out of tests.
• You still check the work. A better model is still a model. Read names, numbers and dates before anything goes to a customer.

Who can use it now

Checked on 7 October 2026 on Cursor’s pricing and plans pages. Prices are per month, before tax.

Plan	Price	What it means for work
Cursor Pro	$20 a month	Grok Bot included, with weekly usage below Pro+
Cursor Pro+	$60 a month	Generous weekly usage; Cursor recommends it for daily agent users
Cursor Ultra	$200 a month	The highest weekly usage
Cursor Teams	$40 per user a month	Every member gets access, with team billing and team-wide Privacy Mode
SuperGrok, SuperGrok Plus, SuperGrok Heavy or X Premium+	Your existing subscription	Link it to get Grok Bot usage; SuperGrok Lite is not included
Cursor Enterprise	Custom	Enterprise access is still rolling out; ask your Cursor account team

Grok Bot runs on Mac, Windows, Linux, iPhone, iPad and Android.

Quick questions

Can I choose which AI model Grok Bot uses?

No. There is no model picker; model selection is managed for you. Enterprise teams can set a model allowlist, but the docs say Grok Bot may not always follow it.

Does Grok Bot use Claude Opus 5.5 right now?

Musk said on 7 October 2026 that it will, “going forward”. The official docs don’t give a date or a task list yet, so ask your Bot what it can do today.

How much does Grok Bot cost?

It is included with Cursor Pro ($20 a month), Pro+ ($60), Ultra ($200) and Teams ($40 per user), or through a linked SuperGrok or X Premium+ subscription. Each includes weekly usage.

Is my work data used to train the models?

With Privacy Mode on, the docs say customer data is not used for training. Training opt-out follows your Cursor account and privacy settings.

When model routing is sorted and the next bottleneck is clicking through desktop apps and local files, see our guide to the best AI agent for desktop workflows in the office.

For a chat-only desk win after the call, see how to use ChatGPT to summarize meeting notes.

For the empty reply box after the call, how to use ChatGPT to draft reply emails faster.

For a long PDF or report that still needs a short brief and owners, see how to use Claude AI to summarize a long report and extract action items.

Sources

• Elon Musk’s post on X, 7 October 2026
• Grok Bot overview (official docs)
• Grok Bot security: models and data (official docs)
• Grok Bot settings: no model picker (official docs)
• Grok Bot FAQ: cost and data (official docs)
• Grok Bot plans and billing (Cursor help)
• Cursor pricing
• Lookonchain report on the announcement