Gemini 4 Neon: Which Gemini 4 Model Should Australians Use?

Tangled strands of glowing cyan and blue optical fibre lights curving around dark metal arches against a dark background.

Gemini 4 Neon is Google’s newest Gemini 4 model. It launched on 18 November 2026 (US time), the third Gemini 4 model Google has made widely available this month after Carbon and Hydrogen, and the fourth Gemini 4 name since Argon in October. If you’re not sure which one your business should be using, you’re not alone. Here’s a plain guide to what Neon is, how it fits between Hydrogen and Carbon, what it costs, where your data is processed, and how to choose.

What Google released

Google introduced Gemini 4 Neon as a model for “complex software engineering, knowledge work, and science workflows” that combines strong reasoning with faster task completion. Its model card says Neon is adapted from Gemini 4 Carbon, accepts text, images, audio, video and documents, and has a context window of up to one million tokens.

Google says Neon is rolling out:

  • For developers: in the Gemini API paid tier through Google AI Studio.
  • For organisations: in Gemini for business, via the API or the model drop-down, and in Google Antigravity, Google’s agent-based coding tool.
  • For individuals: in the Gemini app for paid Google AI Pro and Ultra subscribers.

Google also says Neon ranks near the top of several independent leaderboards and recorded its lowest fabrication rate yet in internal testing. Those are Google’s own claims about results it chose to publish. Treat them as a reason to test Neon, not as proof it’s the best model for your work.

Gemini 4 Neon vs Hydrogen vs Carbon: where each fits

Google’s own guide to choosing a Gemini model is the clearest explanation of the family. In short:

  • Gemini 4 Hydrogen: the everyday default. Google says to “start here for most new projects”. It’s the fastest and cheapest of the three, it’s on the free tier, and it suits everyday coding, interactive apps, document pipelines and agents that call tools.
  • Gemini 4 Neon: the middle model. It “sits between Hydrogen and Carbon”, handling harder reasoning than Hydrogen and finishing faster than Carbon. Google suggests it for complex software work, dense knowledge work and science tasks “where Hydrogen isn’t enough and you don’t need Carbon’s extra verification”. It’s paid tier only.
  • Gemini 4 Carbon: the heavy lifter. Google’s strongest reasoning model spends more thinking per turn to test ideas and check its own work. Google suggests it for system architecture, hard debugging, dense analysis and security reviews, or when a faster model gets stuck.

What about Gemini 4 Argon, which we covered in our Gemini 4 Argon guide in October? Argon is still listed in Google’s API pricing, but it’s no longer one of the three models in Google’s choosing guide. 9to5Google reports that Neon replaces Argon in the Gemini app’s model picker, which now offers Auto, Hydrogen (“Everyday help”), Neon (“Deeper analysis”) and Carbon (“Code & complex work”).

What Gemini 4 Neon costs

On the Gemini API, prices are per million tokens in US dollars, according to Google’s pricing page:

  • Neon: US$3 input and US$15 output. There’s no introductory price, and no free tier.
  • Hydrogen: US$1.50 input and US$7.50 output on the paid tier, with a free tier as well.
  • Carbon: US$2 input and US$10 output on introductory pricing until 30 November 2026, then US$4 input and US$20 output from 1 December.

Two details matter for budgeting. Output prices include “thinking” tokens, so a model that reasons longer costs more per answer. And from December, Carbon’s listed prices sit above Neon’s, so anyone who picked Carbon during its promotional pricing should re-check their costs. Per-token prices also aren’t the whole story: a model that finishes a task in fewer steps can cost less per task even at a higher price per token. The only way to know is to measure on your own work.

Where your data is processed: the Australian question

For Australian organisations that build on Google Cloud, this is the part to read carefully. Google’s Gemini 4 Neon documentation lists it on the global endpoint and the US and EU multi-regions. Google’s data residency page (last updated 19 November 2026, UTC) shows machine learning processing for Neon, Hydrogen and Carbon in the US and EU multi-regions only, not Sydney (australia-southeast1).

The same page warns that requests sent to a global endpoint “may be processed in any Google Cloud location around the world” and so “don’t provide any data residency guarantees”. Among Google’s own models, Gemini 3.5 Flash is the one listed with Australian processing on that page.

If your contracts, clients or sector rules require data to be processed in Australia, no Gemini 4 model is the default answer yet. Our guide to private LLM options in Australia covers the alternatives.

One more line to draw: Google’s pricing page says content sent on the free tier is used to improve its products, while paid tier content isn’t. Staff trying a free tool with client files is a different risk from a billed, governed project.

Which Gemini 4 model should an Australian business use?

A simple way to decide:

  • Staff using the Gemini app for everyday questions, emails and summaries: Auto or Hydrogen will handle most of it. Neon is worth trying for longer reports, spreadsheet analysis and research briefs, if you already pay for Google AI Pro or Ultra.
  • Building a chatbot, document workflow or agent on the API: start with Hydrogen, as Google recommends, and move a task up to Neon only when Hydrogen’s answers aren’t good enough.
  • Hard, high-stakes technical work such as architecture decisions or security reviews: test Carbon, and keep a person checking the output.
  • Still on Gemini 3.1 Pro Preview? Google’s guide names Neon as the recommended upgrade, or Carbon for the hardest problems.

Before you switch models, Google suggests tuning the “thinking level” first. Neon has four levels, from low to extra high. A lower level is faster and cheaper, and it’s often enough.

If you’re choosing tools for a small team rather than building software, our free Choose AI Tools for Business course and the AI tools for small business course are a good place to start.

Using more than one model in the same workflow

You don’t have to pick a single model. Google’s guide says Carbon, Neon and Hydrogen share the same request format, context window and tools, so each step of a workflow can go to the model that suits it. In Google Antigravity, a feature called Orchestrator can now pair Neon as the planner with Hydrogen doing routine steps, handing back to Neon if Hydrogen gets stuck.

That planner-and-worker pattern is useful, but it adds moving parts. Our post on single agent vs multi-agent AI explains why one well-built agent is usually the better place to start.

How to trial Gemini 4 Neon safely

  1. Pick three to five real tasks your team does often, with examples of what a good answer looks like.
  2. Run them on Hydrogen and Neon, at more than one thinking level, and compare quality, speed and cost per task.
  3. Use a paid, billed account for anything involving client or staff information, and check where it’s processed.
  4. Keep API keys locked down and set spending limits. Our post on LLMjacking and stolen AI API keys explains why.
  5. Keep a person reviewing anything that goes to a customer or into a decision.

Caveats

  • Vendor claims. Benchmark rankings and fabrication results come from Google’s own announcement. Independent, hands-on testing of Neon is still early.
  • Fast-moving names. Google has released several Gemini 4 models in weeks, and the app’s model picker has already changed. Check Google’s model guide before you lock anything in.
  • Prices and regions change. Pricing, promotional periods and data residency tables are updated often. The prices here are in US dollars, before currency conversion and any GST.
  • No guaranteed results. Whether a newer model saves time or money depends on your tasks, data and how carefully you test it.

This is general information based on Google’s Gemini 4 Neon announcement, model card, Gemini API and Google Cloud documentation, and 9to5Google’s report, as at 20 November 2026. AI products and pricing change often. It is not legal, privacy or procurement advice for your specific situation.

Not sure which AI model fits your business?

Start with our free AI tools for small business course or the AI agents course. Our AI business audit shows which tasks are worth automating first. Aideveloper builds custom AI software for Australian businesses, choosing and testing the model for each job, with data location, costs and human review planned in. Talk to us about your project.

Featured image: Photo by Georgie Devlin on Pexels.