Trying Gemini 4 Neon: frontier speed, on day one

Things I'm Trying

Trying Gemini 4 Neon: frontier speed, on day one

Google launched Gemini 4 Neon today, November 18, 2026, and it is already in the Gemini app for Pro subscribers. It is the fourth Gemini 4 model, built for complex work at high frequency. I tried it tonight on AuditEase, and the work landed. Here is what Google says it can do, and what I actually saw.

What we cover: - What Gemini 4 Neon is, and where it sits in the lineup - The numbers Google is leading with - How to get it, and what it costs - What I tried tonight on AuditEase - What to watch on day one

Google launched a new model today, and I spent tonight with it. Gemini 4 Neon is the fourth Gemini 4 model in about seven weeks, and it is the one aimed at the work I actually do: complex coding, done fast, over and over. I pointed it at AuditEase, the work landed, and I liked what I saw. I have not had a chance to play with it a ton, so this is night one plus the launch docs, not a review. And yes, the name invites confusion: this Neon is a model, not the database.

What is Gemini 4 Neon?

Gemini 4 Neon is Google's new mid-tier frontier model, announced November 18, 2026 by Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google. It joins a fast-growing family: Argon in October for deep reasoning, Carbon on November 10 for the hardest long-horizon projects, Hydrogen on November 12 for fast everyday work, and now Neon for high-frequency complex work.

Google's pitch is "frontier-level intelligence with fast end-to-end task completion" across software engineering, knowledge work, and science. In plain terms, it is supposed to be the model that does hard jobs without making you wait for them.

The specs: a 1 million-token context window, four thinking levels (low, medium, high by default, and xhigh), and the model ID gemini-4-neon. It is built to run alone or to coordinate parallel Hydrogen subagents in Antigravity, Google's agentic coding platform. In the Gemini app, Neon now holds the "Deeper analysis" slot. Argon is still a model, still in the API, AI Studio, and Antigravity, it just is not in the app picker anymore.

The numbers Google is leading with

Launch day is a benchmark parade, so here are the numbers Google puts first, with the caveat that they are Google's figures and independent tracking was incomplete at launch:

  • 61 on the Artificial Analysis Intelligence Index at xhigh, third overall, just behind larger, slower models, and ahead of Claude Opus 5.5 and GPT-6.2 Sol on the same index.
  • First on the Artificial Analysis Coding Agent Index at 58.6.
  • Second on the Vals Index at 70.9 percent, and second on steerability in the Agent Arena Index.
  • 65.4 percent on CursorBench 4.1.
  • The lowest fabrication rate yet in Google's internal Antigravity testing.
  • Glean's early evaluation: it matched or exceeded Carbon on overall quality at 3x the speed and 60 percent of the cost.

The coding agent index is the number I care about most, because that is the job I gave it tonight. First place on day one is a claim worth testing, not a verdict.

How to get it, and what it costs

Neon is rolling out today in the Gemini API on the paid tier in Google AI Studio and Android Studio, in Gemini for business through the API and the model dropdown, in Google Antigravity for paid Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers. It is not available on the free tier.

Pricing is $3 input and $15 output per 1M tokens, with cached input at $0.15, a 95 percent discount. For price context, Carbon's promotional rates of $2 and $10 run through November 30, then rise to $4 and $20, which makes Neon 25 percent cheaper per token from December 1. Hydrogen is $1.50 and $7.50.

A cyber variant, Gemini 4 Neon for cyber, is rolling out to trusted defenders in the Fairwind Program over the coming days.

What I tried tonight

I tried it on AuditEase tonight, working through T3, which runs on Antigravity, with my Google AI Pro subscription. The night's work was a feature build: per-project branded reports, a scoring update, and a report UI rework. The usage felt good, and the model felt smart about the code instead of guessing at it. It was working really well, and I am really happy with how good it is doing so far.

One night is not a verdict. But the speed is the thing I noticed first: it does the work and finishes it, instead of making me wait while it thinks about doing the work.

What to watch

A few honest caveats before anyone switches:

  • Launch-day numbers are Google's own, and independent verification is thin so far. The coding agent index lead is worth a week of real use, not a retweet.
  • The naming is confusing. Four models in seven weeks, all named after elements. Argon is still around, still the deep long-horizon pick, while Neon is the fast one for frequent complex work. They are built for different jobs, so the lineup is a toolbox, not a ranking.
  • There is no free tier, so the value case depends on whether the speed holds up on your work, not on the benchmarks.
  • I have one night of use. The pull requests landed, but I have not tested a failed task, a big refactor, or a long multi-step job yet. Those are the next three things to try.

Final thoughts

Night one verdict: it is fast, it is smart about code, and it is the model I will reach for first tomorrow. Next I want to give it a deliberately messy refactor and see how it holds up, because that is the test that tells me whether I can trust it.

Read next: Five AI tools I'm actually using every day

/things-im-trying/2026-11/trying-gemini-4-neon-frontier-speed-day-one austin.amento.dev