It is not worth paying extra to run it on everyday email replies. If you already pay for Google Workspace, Google says premium AI is included in Google Workspace plans, including a Gemini AI assistant in Gmail and Docs. The real question is which tasks deserve the bigger model, and what each of those tasks costs.

This guide covers what Neon adds, which Gmail and Docs jobs actually need it, how to work out the cost per task, and when a short spend review pays for itself.

What Google actually released this week

Google released Gemini 4 Neon on November 18, priced at $3 per million input tokens and $15 per million output tokens. According to Google's announcement, it is rolling out in the Gemini API in Google AI Studio, in Gemini for business (via API and model dropdown), in Google Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers.

In plain terms, think of it as the model for heavy reading jobs. 9to5Google's launch coverage says it reads uploads like spreadsheets, PDFs and photos more precisely, and it pulls in relevant details from Google apps like Gmail and Docs. It has a 1 million-token context window with four thinking levels, the same as Hydrogen. That means you can hand it a very large pile of material at once: a full contract set, a year of vendor emails, a long policy manual.

It also costs more than the lighter option. Neon is twice as expensive as Hydrogen. So upgrading is not one decision. It is a decision about which model each kind of task should use.

$3
Neon price per 1M input tokens
$15
Neon price per 1M output tokens
$0.15
Cached input price per 1M tokens
1M
Token context window

The question owners are really asking

When a clinic manager, a CPA or the owner of an HVAC shop asks "should we upgrade?", they usually mean one of three things:

  1. "Are we already paying for this?" You may be, at least in part, so check before you buy.
  2. "Will it save my team time on the work that actually eats our week?" That depends on the job.
  3. "Could this quietly run up a bill nobody is watching?" It can, especially if anyone on your team or a vendor you use is calling the API.

Let's take them in order.

Are you already paying for Gemini?

Premium AI is included in Google Workspace plans, with a Gemini AI assistant in Gmail, Docs, Meet and more. Google stopped charging for the separate Gemini for Workspace add-ons after January 31, 2025, because the AI features were folded into the Business and Enterprise plans. For existing subscriptions with more than 10 user licenses, the related price changes took effect starting March 17, 2025.

If your office is on one of those plans, check whether your team is already using the assistant in Gmail and Docs. We wouldn't start by paying more per seat. Start by checking what your team actually uses, and whether anyone is also paying for a separate AI tool on a company card.

Tip

Before you buy anything, open your Google Workspace admin billing page and your last three card statements. Look for any AI subscription that does the same thing as the assistant already in your plan. In our view, overlapping tools are an easy place for AI spending to leak.

Where Neon shows up for a business

Neon reaches a business in two main ways. First, inside Gemini for business, through the model dropdown. Second, through the Gemini API in Google AI Studio, where Google says Neon is rolling out; custom tools and automations typically reach Google's models this way. For the first, check your plan and settings. The second is billed by the token, so it is where costs can climb without anyone noticing.

Which Gmail and Docs jobs actually need Neon

Our rule of thumb: pay for the bigger model when the job has a lot to read and the answer has to be right. Short, routine writing does not need it.

JobNeeds Neon?Why
Drafting a two-line email replyNoShort input, short output. A lighter model should do.
Summarizing one email threadUsually noUnless the thread is long and the stakes are high.
Pulling key terms out of a long lease or vendor contractYesLong document, and a missed clause costs money.
Turning a month of job-site photos and invoices into a client reportYesMixed files: photos, PDFs, spreadsheets.
Reconciling a spreadsheet against emailed receiptsYesDetail-heavy, and it pulls from Gmail and Sheets together.
Rewriting a Doc in a friendlier toneNoStyle work. Any model handles it.
Building an intake summary from a stack of patient or client formsOftenVolume plus accuracy. Keep privacy rules in mind (see pitfalls).
Weekly "what happened" brief from team emailsSometimesWorth testing both models side by side.

In our view, the jobs in the "yes" rows are where Neon earns its price. It reads spreadsheets, PDFs and photos and pulls details from Gmail and Docs. For an accounting office at tax time, a law office in discovery, or a property manager juggling leases, we'd treat that as real work, not a demo.

What it costs per task (with the math shown)

In our experience, per-seat pricing hides the real driver of AI cost: tokens per task times tasks per month. A token is a small chunk of text, roughly part of a word. You pay for what goes in (input) and what comes out (output), and output costs more.

The rates, from launch coverage:

Worked example

Hypothetical: a 12-person accounting office runs a document-review automation through the API. Each month it handles 400 client packets. Each packet is about 30,000 tokens of input (a few PDFs and an email thread) and produces a 2,000-token summary.

That is cheap compared with staff time. Now look at how the same setup goes wrong.

Hypothetical, poorly built: the automation re-sends a 200,000-token reference binder with every packet, asks for a long write-up "just in case", and retries failed runs three times.

Same office, same job, more than six times the bill. The fix here isn't a cheaper plan. It's a better-built workflow. Caching that reference binder drops its cost from $3 to $0.15 per million tokens, and trimming what goes into each request cuts the rest. This is the kind of thing we fix in context engineering and token optimization.

Neon price per 1M tokens (USD)
03.757.511.25153Input0.15Cached input15Output

A quick way to estimate your own number

  1. Pick one job. For example, "summarize each new lease."
  2. Count how often it runs. Per week or per month.
  3. Estimate the size. As a rough guide, a page of plain text is a few hundred tokens. A long PDF can run to tens of thousands.
  4. Multiply. Input tokens ร— $3 per million, plus output tokens ร— $15 per million.
  5. Compare it with the time saved. If the job takes a staff member a good while and the model costs very little, the math is easy. If the model's draft takes just as long to check, it's not.
Heads up

Launch prices for new models often change. Before you budget a full year, check the current rate on the provider's own pricing page and look for promotional pricing that has an end date. Build your estimate so you can change one number and recompute.

Pitfalls we'd watch for

Paying twice. You may have Gemini in your Workspace plan, a separate chatbot subscription for a few staff, and a vendor tool that runs its own AI on top. Three bills for one job.

The silent API meter. Seat plans have a fixed price. API usage does not. A connector, a Zapier-style automation, or a developer's test script can run all weekend. If nobody sets a budget alert, the first warning may be the invoice.

Using the big model for small work. Neon costs twice what Hydrogen does. Sending every "draft a polite reply" request to the bigger model is like hiring a senior attorney to file paperwork.

Long outputs nobody reads. Output tokens cost five times as much as input tokens at these rates. Ask for the short version: the five key dates, the three red flags, not a ten-page essay.

Sensitive data. If you run a clinic, law firm or accounting office, treat client information as sensitive. Before pointing any AI tool at patient files, client tax records or privileged email, confirm which plan and settings you're on and ask what your obligations are. That's a compliance question for your advisor, not a setting to guess at.

Never measuring the "before." If you don't know how long a task took before AI, you can't tell whether the upgrade paid off.

When a 30-minute spend audit pays for itself

A short spend review makes sense when any of these are true:

In a short review you can usually answer these questions:

  1. What are we paying for? Seats, subscriptions, API usage.
  2. What's overlapping? Tools that do the same job.
  3. Which tasks use the most tokens? Usually one or two automations.
  4. Which model should each task use? Big model for heavy reading, lighter model for routine writing.

Here's the simple test for whether it's worth your time. Take the worked example above. The gap between the well-built and poorly-built versions was a couple of hundred dollars a month for a single workflow. If your review finds even one leak like that, it pays for the half hour many times over in the first month. If it finds nothing, you leave with a clear picture and a budget alert set, which is also worth having before you upgrade.

Our token expenditure audits do exactly this. We map where your AI spend goes, which jobs need a bigger model, and where caching or tighter prompts cut the bill. We don't sell you another subscription.

What we'd do this week if we were you

Here's our recommendation for a typical Southern California office, whether that's a dental group in the Inland Empire, a law office in Orange County, or a trades company covering the 909 and 951:

  1. Don't upgrade the whole office on day one. In our view, most everyday Gmail and Docs work doesn't need the bigger model.
  2. Confirm what your Workspace plan already includes. Check which Gemini features your current Workspace plan includes before you buy anything extra.
  3. List your three heaviest reading jobs. Contracts, intake packets, monthly reports, spreadsheet reconciliations.
  4. Test the bigger model on those jobs only. Run the same five real examples through the lighter model and through Neon. Compare accuracy and checking time, not just speed.
  5. Price each job per task using the formula above, so the decision is about dollars per result.
  6. Set a budget alert on any API billing before you let an automation run unattended.
  7. Review it after the trial. Keep what saved time. Drop what didn't.

If those heavy jobs are worth automating end to end, so that documents come in, summaries go out and a person checks the flagged items, that's an AI automation project, and we think that's where per-task math matters most. We work with businesses across Southern California on exactly this kind of setup.

Checklist: is Gemini 4 Neon worth it for your business?

If you can tick most of these, you're probably ready to make the upgrade decision yourself. If you can't, that's normal two days after a launch, and it's what a short review is for.

Next step

Want a second set of eyes before you spend more on AI? Book a Free Automation Audit. We'll look at where your AI money goes today, which Gmail and Docs jobs actually need Gemini 4 Neon, and what each one should cost per task. Plain answers, no upsell.

Sources