Gemini 4 Neon Explained: Google's Middle Model — and Why It Only Makes Sense from December 1

N
Navs
Published on November 22, 20267 min read
Tags:Gemini 4 NeonGoogleAI modelsGeminicoding agentsAI pricing
Gemini 4 Neon Explained: Google's Middle Model — and Why It Only Makes Sense from December 1

Google has now shipped four Gemini 4 models in about seven weeks, and the newest one is the hardest to place. Gemini 4 Neon, announced on November 18, 2026, sits between Hydrogen, the fast everyday model, and Carbon, the heavyweight built for the hardest long-horizon work. Google's pitch is that Neon brings frontier-level intelligence to complex software engineering, knowledge work, and science workflows — and finishes those jobs fast.

That pitch comes with a pricing twist that decides whether Neon makes sense for you right now. Neon costs $3 per million input tokens and $15 per million output tokens. Carbon, the model above it, still costs $2 and $10 until November 30 — which makes Neon the more expensive model today, and the cheaper one from December 1, when Carbon's introductory pricing ends and its rates double to $4 and $20. This guide explains what Neon is for, what Google claims it can do, where you can use it, and who should switch now, wait two weeks, or stay put.

Neon, Hydrogen, and Carbon: three models, three jobs

Each tier is defined by a job, not just a price. Hydrogen handles fast everyday execution. Carbon takes on the hardest, long-horizon projects and deep research. Neon is the middle: complex work that runs often, where quality matters but waiting for Carbon's long deliberation does not pay.

ModelJobInput / output per 1M tokens
Gemini 4 HydrogenFast everyday execution$1.50 / $7.50
Gemini 4 NeonComplex coding, knowledge work, science$3.00 / $15.00
Gemini 4 CarbonHardest, long-horizon work$2.00 / $10.00 until Nov 30; $4.00 / $20.00 from Dec 1

Neon is adapted from Carbon, according to Google's model card, which explains why its scores land so close to the flagship. The difference is the operating point: Neon trades some of Carbon's deliberation for speed, so it can run high-frequency workflows — the coding agent that works all day, the document pipeline that never stops — without the flagship's latency or, from December, its price. In Google Antigravity, it can also act as a fast Orchestrator directing parallel Hydrogen subagents, sharing progress as it works and handling context compaction in long sessions.

All three models share a 1 million-token unified context window (1,048,576 tokens, covering input, thinking, and output), and all three expose four thinking levels. That window matters for Neon's target work: enterprise jobs live in spreadsheets, PDFs, charts, audio, and video, and Neon can synthesize across all of them and turn the result into structured deliverables.

The benchmark case Google is making

Every score in this section is vendor-reported, taken from Google's launch announcement and the independent indexes it cites. Navs has not independently tested Neon. Treat them as Google's case for the model, not as settled fact.

The headline number is a 61 on the Artificial Analysis Intelligence Index at Neon's xhigh thinking level, ranking third overall — behind Claude Fable 5.2 and GPT-6.2 Astra, and ahead of Claude Opus 5.5 at 59 and GPT-6.2 Sol at 57. Two details keep that number honest: 61 is the xhigh score, while at the default high level Neon scores 60, according to Artificial Analysis data — the difference between the setting developers actually use and the setting Google leads with.

Coding is where Neon makes its strongest claim. It ranks first overall on the Artificial Analysis Coding Agent Index at 58.6, scores 65.2% on Terminal-Bench 4.1, and 46.2% on FrontierCode 2.0 at its default high setting. Google says Neon handles multi-file debugging and refactoring across repositories with targeted, surgical edits.

Google's benchmark table comparing Gemini 4 Neon with Claude Opus 5.5 and GPT-6.2 Sol across intelligence, coding, and science tests

The enterprise and science results follow the same pattern. Neon ranks second on the Vals Index at 70.9%, first on both the Legal Research Bench and the Tax Agent Bench, and first on most Surge AI benchmarks across finance, accounting, and spreadsheet modeling. It debuts second globally on the Agent Arena Index at 1336, and second on steerability — how cleanly a model adapts to mid-task corrections. It ranks first on Zapier's AutomationBench at 56.3% and on Andon Labs' Vending-Bench 2 at $16,843. In science it leads Terminal-Bench Science at 70.5%, scores 73.5% on Chartography, and 75.8% on Agents' Last Exam v1.1.

Google also pushes a security angle: Neon produces more secure code than Carbon on AutoBaxBench, reaching state-of-the-art on the hard tier at 76%, and a dedicated Neon for cyber edition is rolling out to trusted defenders through the Fairwind Program.

The honest reading: Neon is a genuinely strong coding model by every index Google chose to publish — and Google chose the indexes. The Coding Agent Index lead is the result to watch, because it is the one most directly tied to the work Neon is sold for.

Thinking levels: four settings, no off switch

Neon ships with four thinking levels — low, medium, high, and xhigh — with high as the default, letting teams dial intelligence and token spend to what each workload needs. Xhigh adds deeper, multi-step reasoning for the hardest work, and it is where Neon records its highest intelligence score.

Two things affect the bill. First, thinking tokens are billed at the output rate, so a model that reasons more costs more per call. Second, thinking cannot be switched off: Neon has no minimal or off setting, so every call pays for some reasoning. Artificial Analysis data puts Neon's cost per Intelligence Index task at $3.74 at high and $6.44 at xhigh — the extra point of intelligence at xhigh costs about 70% more per task. For routine work, high is the sensible default; xhigh is for jobs where a wrong answer costs more than the extra tokens.

What it costs, and the December 1 price flip

On the Gemini API's Standard tier, Neon costs $3.00 per million input tokens, $15.00 per million output tokens, and $0.15 per million cached input tokens — a 95% discount through implicit caching. There is no free tier: Neon is paid API only.

For work that can wait, Batch and Flex halve every rate, to $1.50 input and $7.50 output per million tokens — exactly Hydrogen's standard price. Priority costs $5.40 and $27. For nightly code review, eval runs, and document backfills, Batch or Flex turns Neon into a Hydrogen-priced model with a stronger brain.

Artificial Analysis cost-per-task chart showing Gemini 4 Neon against competing frontier models at high thinking

The December 1 flip is the part to get right. Carbon's $2 and $10 rates are introductory and expire on November 30, 2026. From December 1, Carbon costs $4 and $20 — double. Neon, at a flat $3 and $15, goes from costing 1.5 times Carbon per token to costing 25% less per token. In cost per task, Neon already wins today: Google claims, citing Artificial Analysis v5.0.1 data, that Neon costs 32% less per task than Carbon at its default high level, despite the higher per-token rate, because it spends fewer tokens finishing the same work. The claim is Google's, built on third-party data — credible direction, not a guarantee for your workload.

Where you can use Neon today

Neon began rolling out on November 18, 2026, in the Gemini API paid tier in Google AI Studio, in Gemini for business via the 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. Google AI Pro costs $19.99 per month in the US.

Gemini homepage with the headline Supercharge your creativity and productivity.

In the Gemini app, Neon targets work that needs frontier-level reasoning and fast responses: higher-quality writing, research that pulls in current information from across the web, and multi-step jobs like analyzing billing statements. It reads uploads — spreadsheets, PDFs, photos — more precisely than the models it replaces in the picker, and can pull details from Google apps like Gmail and Docs. Note what is not on the list: Google's announcement names no free tier and no Vertex AI availability, so do not plan a rollout around either until Google says so.

Who should switch now, who should wait, and who should stay on Hydrogen

Switch now if you run coding agents or document pipelines at volume and your bill is driven by cost per finished task rather than cost per token. Neon's case is strongest exactly there: the Coding Agent Index lead, the 32% lower cost per task than Carbon that Google claims, and Batch or Flex pricing that matches Hydrogen's standard rate. If your agent work runs on Carbon's introductory pricing, test Neon on a slice of it this week — the per-task math, not the per-token sticker, is the comparison that matters.

Wait until December 1 if you are choosing on per-token price alone, or if your workload is mostly Standard-tier API calls. On December 1 the comparison flips cleanly: Neon becomes 25% cheaper per token than Carbon at every tier, and the only reason to pay Carbon's rate after that is work that genuinely needs the flagship's deeper deliberation. There is no penalty for waiting nine days.

Stay on Hydrogen if your work is fast, everyday execution — short prompts, single-turn answers, agent steps that do not need deep reasoning. Hydrogen is half Neon's price, the only one of the three with a free tier, and the model Neon itself delegates to inside Antigravity. The family is designed to be mixed: Hydrogen for the steps, Neon for the complex jobs, Carbon for the hardest ones.

Neon is not a revolution, and Google does not present it as one. It is a middle model with a clear job, a strong coding record, and a price that only makes sense once you read the date on Carbon's discount. If your work fits the middle — complex, frequent, and billed by the finished task — try it this month: trial before December 1, switch after it.

Sources

  • Introducing Gemini 4 Neon — blog․google․com
  • Gemini 4 Neon model card — deepmind․google․com
  • Gemini 4 Neon launches: Carbon quality at Flash speeds, for AI Pro — 9to5google․com
  • Artificial Analysis Intelligence Index — artificialanalysis․ai
Share this article