Google Launches Gemini 4 Neon: Pricing, Availability and What Google Claims
Gemini 4 Neon costs $3/$15 per million tokens with a 1M-token context window. What Google confirmed, where to use it, and how to read its benchmarks.
Table of contents
Google has added a fourth model to its Gemini 4 family. Gemini 4 Neon, announced on November 18, 2026, is pitched as near-frontier reasoning at a lower cost per task than Google's top model, Gemini 4 Carbon. It is available in the Gemini API on the paid tier, in Google AI Studio, in Google Antigravity and in the Gemini app for Google AI Pro and Ultra subscribers.
This article sets out what Google has actually confirmed — pricing, context window, availability and positioning — and how to read the long list of benchmark numbers that came with the launch.
The short version
- What it is: a mid-to-high tier model that sits between Gemini 4 Hydrogen (fast, everyday tasks) and Gemini 4 Carbon (the hardest, longest tasks).
- Price (API): $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens.
- Context window: 1 million tokens.
- Reasoning control: four thinking levels, with "high" as the default and "xhigh" as the deepest setting.
- Where: Gemini API (paid tier) via Google AI Studio and Android Studio, Gemini for business and Gemini Enterprise, Google Antigravity, and the Gemini app for AI Pro and Ultra subscribers.
All of the above comes from Google's launch post, written by Koray Kavukcuoglu of Google DeepMind.
Where Neon fits in the Gemini 4 family
Google now describes the Gemini 4 line-up by job rather than by size. In its announcement, Carbon handles the most demanding long-horizon projects, Hydrogen handles fast everyday execution, and Neon targets what Google calls complex, high-frequency workflows — the kind of coding, analysis and enterprise automation that runs many times a day and where cost per task matters.
In the consumer Gemini app, 9to5Google reports that the model picker now maps to these roles: Hydrogen for everyday help, Neon for deeper analysis and Carbon for code and complex work. According to the same report, Neon replaces the earlier Gemini 4 Argon model in the app, and Gemini 3.1 Pro is being retired.
If you have been using Argon for coding and analysis inside the Gemini app, Neon is effectively its successor there.
Pricing in context
Neon's list price is $3 per million input tokens and $15 per million output tokens, with cached input at $0.15 per million — a 95% discount that matters a lot for agentic workloads which resend the same large context (a code repository, a long contract, a spreadsheet) on every step.
Google's headline cost claim is about cost per task, not price per token. A model that reasons for longer can be cheaper per token yet more expensive per completed task. Google says that at its default "high" setting, Neon costs 32% less per task than Carbon, and that it delivers roughly 50% to 80% lower cost per task than top competing frontier models on two third-party leaderboards it cites (Vals Index and Agent Arena).
9to5Google also notes a timing detail worth watching: Argon and Carbon are on promotional pricing of $2 input / $10 output per million tokens until November 30, after which they rise to $4 / $20. If you are comparing models this month, compare against post-promotion prices, or your estimates will be out of date in two weeks.
Thinking levels: the setting that changes your bill
Like Hydrogen, Neon exposes four thinking levels. Google says the default is "high", and that "xhigh" adds deeper multi-step reasoning for the hardest workloads and produces Neon's highest scores.
The practical takeaway for developers: benchmark scores and costs are tied to a thinking level. Several of Google's quoted results are measured at xhigh, while the 32% cost advantage over Carbon is quoted at the default high setting. When you run your own evaluation, record which level you used for both quality and cost — otherwise you are comparing different configurations.
How to read Google's benchmark claims
Google's post lists more than twenty results. A few of the most prominent, as stated by Google:
| Claim (as reported by Google) | Result |
|---|---|
| Artificial Analysis Intelligence Index (xhigh) | 61, ranked #3 overall |
| Artificial Analysis Coding Agent Index | #1 overall at 58.6 |
| Terminal-Bench 4.1 | 65.2% |
| Vals Index | #2 at 70.9% |
| Zapier AutomationBench | #1 overall at 56.3% |
| HLE-Diamond | 71.3% |
Some of these leaderboards are run by third parties, but the numbers above are reproduced from Google's own announcement. KernelBrief has not independently verified them, and early-launch rankings on public leaderboards often move as more models and more runs are added.
Three things are worth keeping in mind:
- Leaderboard rank is not your workload. A #1 result on an automation benchmark does not guarantee better results on your codebase, your documents or your prompts.
- Configuration matters. As noted above, results differ by thinking level, and agentic results also depend heavily on the harness — Google itself reports one coding result in a "mini" harness and a much higher one inside its own Antigravity tool.
- Cost and quality must be measured together. The interesting claim here is not raw intelligence; it is whether Neon gets close to Carbon's quality for meaningfully less money per completed task.
Safety claims
Google says Neon is built on the same safety and security stack as Carbon and maintains comparable protections against severe misuse and prompt injection. Among the figures it reports: a 0.09% misaligned-action rate on its agentic safety suite and, on Gray Swan's indirect prompt-injection benchmark, no successful attacks at Neon's highest thinking level. Google points to the Gemini 4 Neon model card for full details. As with the capability numbers, these are the company's own reported evaluations.
What this means for developers and teams
If you build on the Gemini API: Neon is a natural candidate wherever you currently route complex tasks to Carbon purely for quality. A sensible approach is a side-by-side evaluation on a sample of real tasks, measuring task success rate and total cost per task at the default thinking level, then at xhigh only for the hardest slice.
If you use the Gemini app: AI Pro and Ultra subscribers should see Neon in the model picker as the "deeper analysis" option. Free-tier availability was not announced.
If you run agents: the cached-input price and the per-task cost claims are aimed squarely at you. Long-running agents that repeatedly send large contexts benefit most from caching, so check that your integration actually uses context caching before drawing conclusions about cost.
If you are new to working with large language models and want a grounding in the security side of agent deployments, our guide to passkeys and phishing-resistant sign-in covers a related principle: reduce what an attacker — or a misbehaving tool — can do with stolen access.
What we still don't know
- Independent evaluations. Third-party testing at scale takes days to weeks. Expect leaderboard positions to shift.
- Free-tier access. Google's post lists paid surfaces; it does not announce Neon for free Gemini users.
- Vertex AI availability. Google's launch post does not mention Vertex AI, so check the Google Cloud documentation before planning a Vertex deployment.
We will update this article if Google publishes further details or if independent benchmarks materially change the picture.
Frequently asked questions
+When was Gemini 4 Neon released?
Google announced Gemini 4 Neon on November 18, 2026 (US time) and began rolling it out the same day.
+How much does Gemini 4 Neon cost in the API?
Google lists $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens on the paid tier of the Gemini API.
+What is Gemini 4 Neon's context window?
Google says Neon supports a context window of 1 million tokens.
+Is Gemini 4 Neon better than Gemini 4 Carbon?
Not overall. Google positions Carbon for the hardest long-horizon work and Neon as a faster, cheaper option for complex but high-frequency tasks. Google says Neon's default setting costs 32% less per task than Carbon. Whether it matches Carbon on your workload is something to test yourself.
Sources & references
- Google — Introducing Gemini 4 Neon (Koray Kavukcuoglu, Nov 18, 2026)
- 9to5Google — Google launches Gemini 4 Neon, live for AI Pro (Nov 18, 2026)
- Google for Developers on X — Gemini 4 Neon announcement
Spotted an error? See our Corrections Policy.
- #gemini
- #llm
- #ai-models
- #generative-ai
The KernelBrief editorial desk researches, writes and fact-checks every article published on this site. Every article is checked against primary sources (official documentation, vendor announcements, standards bodies and original research) before publication. Every factual claim is linked to its source, and corrections are handled under our published Corrections Policy.
Related articles
How Large Language Models Work: Tokens, Context Windows and Parameters Explained
How large language models work, in plain terms: what tokens are, how context windows fill up, what parameters mean, and how to count tokens in code.
· 10 min read
F5 BIG-IP Buffer Overflow Flaws: Six Critical CVEs, Who Is Affected and How to Patch
F5 disclosed six critical BIG-IP stack buffer overflows (CVSS 9.8) that crash TMM without login. Affected versions, risky configs and hotfixes.
· 4 min read
VMware vCenter Zero-Day CVE-2026-123890 Was Exploited for Months: Who Is Affected and How to Patch
VMSA-2026-0009 fixes vCenter auth bypass CVE-2026-123890, exploited since September per Mandiant. Fixed builds, no workaround, and what to check.
· 6 min read