Skip to content
Chain AI Wire
AI Models

Google Launches Gemini 4 Neon: Pricing, Benchmarks and Where It Sits in the Gemini 4 Family

Google's Gemini 4 Neon costs $3 input and $15 output per million tokens, reads one million tokens of context, and sits between Carbon and Hydrogen — with an Artificial Analysis score of 61.

Google Gemini 4 Neon launch - four neon flames in violet, orange, blue and red on a dark laboratory bench
Google Gemini 4 Neon brings frontier-level intelligence with fast end-to-end task completion. Image: Chain AI Wire.

At a glance: Google introduced Gemini 4 Neon on 18 November 2026, a new middle tier in the Gemini 4 family that sits between Carbon and Hydrogen. It costs $3 per million input tokens and $15 per million output tokens, with cached input at $0.15, and it reads up to one million tokens of context. Google says Neon delivers frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work and science workflows, and that at its default High thinking level it costs 32% less per finished task than Carbon. The benchmark figures below are Google’s own, published in its launch announcement, and independent testing has not yet had time to check them.

Google has added a third model to its Gemini 4 lineup, and the timing says a lot about where the race has moved. Gemini 4 Carbon arrived last week as the heavyweight for the hardest long-horizon projects, and Hydrogen as the fast everyday model. Neon is the one in between, built for work that is too demanding for a light model and too frequent to keep running on the flagship. Google announced Gemini 4 Neon on 18 November, and it is rolling out the same day in the Gemini API, Google AI Studio, Gemini for business, Google Antigravity and the Gemini app.

Where Neon sits in the Gemini 4 family

Google describes the family in plain terms. Carbon tackles the hardest long-horizon projects. Hydrogen handles fast everyday execution. Neon balances depth and efficiency, delivering frontier-level reasoning for complex, high-frequency workflows, faster and more cost efficiently than the larger models. In practice that means long-running agent work: debugging a large codebase across many files, pulling a research brief together from spreadsheets, PDFs, charts, audio and video, or running a multi-step enterprise analysis that has to finish as a structured deliverable rather than a chat answer.

The context window is one million tokens, the same as Carbon, and Neon can take in text, images, audio and video in a single prompt. Google says it synthesises across all of those inputs and turns them into structured deliverables, which is the part that matters for enterprise teams. A model that can read a whole quarter of invoices, contracts and spreadsheets in one go is a different tool from one that needs the work chopped into pieces.

Pricing: $3 in, $15 out, and cheaper than Carbon after November

The price is straightforward. Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens. Cached input costs $0.15 per million tokens, a 95% discount on the standard input rate, which matters for apps that reuse the same long prompts again and again. That puts Neon at twice the per-token price of Hydrogen, and above Carbon’s introductory rate, but the comparison flips at the end of the month. Carbon’s promotional pricing of $2 input and $10 output runs through 30 November, after which it rises to $4 and $20, and Neon then becomes 25% cheaper per token than Carbon.

ModelInput / 1M tokensOutput / 1M tokensCached input / 1M tokens
Gemini 4 Neon$3$15$0.15
Gemini 4 Carbon (introductory, to 30 Nov)$2$10Not disclosed
Gemini 4 Carbon (from 1 Dec)$4$20Not disclosed
GPT-6.2 Astra$8$40$0.40
Claude Fable 5.2$8$40Not disclosed
Qwen4-Max$1.50$4.50$0.075

List price is only half the story, and Google knows it. The company frames Neon around cost per finished task, not cost per token. At its default High thinking level, Google says Neon pairs a high Artificial Analysis Intelligence Index score with a 32% lower cost per task than Carbon, and that it sits on the Pareto frontier on both the Vals Index and the Agent Arena Index at High, with roughly 50% to 80% lower cost per task than top competing frontier models. Those are Google’s own figures, from its own selected benchmarks, so they should be read as the vendor’s case for the model until outside labs run the same workloads.

Four thinking levels, from Low to Extra High

Neon ships with four thinking levels: Low, Medium, High and Extra High. Google says it has refined the levels with each new model release, so teams can dial intelligence and token spend to what each workload needs, with a smoother progression from one tier to the next. High is the default. Extra High adds deeper, multi-step reasoning for the hardest engineering, analytical and scientific workloads, and it is at Extra High that Neon records its highest intelligence score, while still costing 36% or more less per task than competing models at their highest setting, according to Google.

Benchmarks: third overall on the Artificial Analysis index

On the Artificial Analysis Intelligence Index, Neon scores 61 at its Extra High thinking level, ranking third overall and landing just behind larger, slower models, and just above Carbon’s 60. On the Vals Index it ranks second with 70.9%, and first on both the Legal Research Bench and the Tax Agent Bench. It also ranks second globally on the Agent Arena Index with a score of 1336, across more than 4,200 real-world agent sessions, and second on steerability, which measures how cleanly a model adapts to a correction in the middle of a task.

For coding, Google reports that Neon 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. On SWE-sweep it leads the mini-SWE harness at 12.8%, and Google says the figure more than triples to 39.0% inside Google Antigravity, which is where the model is meant to work as a fast, high-judgment orchestrator directing parallel Hydrogen subagents. For science and maths, Google reports best-in-class results on Terminal-Bench Science at 70.5%, RiemannBench at 84.0% and HLE-Diamond at 71.3%. Again, these are the numbers Google chose to publish. The Artificial Analysis score comes from a third-party evaluator, but the rest are Google’s own reporting of its results.

Built for production code and long agent runs

Neon is engineered for long-running agent workflows, and Google’s launch notes spend most of their space on that rather than on single-turn answers. The model is meant to sustain performance through complex, tool-heavy tasks end to end, whether running standalone or acting as an orchestrator in Antigravity. Google says it cuts down on redundant file reads, shares progress as it works, and handles context compaction reliably in long sessions. For production code, that translates into multi-file debugging and refactoring across repositories with targeted, surgical edits, rather than rewriting whole files.

The same long-horizon pitch shows up in the enterprise benchmarks Google cites. Neon ranks first on most Surge AI benchmarks across finance, accounting, spreadsheet modelling and complex constraint following, scores 73.5% on Chartography for reading dense professional charts, and takes first place overall on both Zapier’s AutomationBench at 56.3% and Andon Labs’ Vending-Bench 2, a long-horizon simulation of running a business, where Google reports $16,843. Those are niche evaluations, but they are the ones that look most like real office work.

A cyber version for trusted defenders

Google is also shipping a cyber version of Neon, and it gets a whole section of the launch. The company says Gemini 4 Neon for cyber produces more secure code than Carbon across easy, medium and hard challenges on AutoBaxBench, reaching state of the art on the hard tier at 76%. On WebExploitBench it reports 51.26% at half the cost of competing large models, and on SEC-bench Pro it reports 90.5% while costing roughly half as much per challenge. Patching is covered too: Neon reaches 70.7% on CWE-bench v2, a 3.4% uplift over Carbon, according to Google. The cyber version is rolling out to trusted defenders in the coming days as part of Google’s Fairwind Program.

Safety is framed as a match for Carbon rather than a step down. Google says Neon maintains Carbon-level protections against severe misuse and prompt injection, with strong real-time safeguards against cyber and biosecurity misuse, while refusing far less legitimate work in those areas. On Gray Swan’s indirect prompt injection benchmark, Google says Neon resisted every attack at its highest thinking level, matching a 0.0% attack success rate. As with the capability figures, these are Google’s own test results, published by the company that built the model.

Where you can use Gemini 4 Neon today

Neon began rolling out on 18 November. Developers can build with it through the Gemini API paid tier in Google AI Studio and Android Studio. Enterprises get it in Gemini for business, through the API or the model dropdown, and in Google Antigravity. Consumers on a paid Google AI Pro or Ultra plan can select it from the model dropdown in the Gemini app, and in Antigravity. The contrast with the rest of the month’s launches is worth noting: GPT-6.2 Astra costs $8 input and $40 output per million tokens, Claude Fable 5.2 the same, and Qwen4-Max $1.50 and $4.50. Neon sits between the cheapest and the most expensive flagships, and Google’s bet is that speed and cost per finished task will matter more than either extreme.

What to watch

  • Independent benchmarks: whether outside labs reproduce the Artificial Analysis ranking and the cost-per-task figures Google reports.
  • Carbon’s price rise on 1 December: when the introductory $2 / $10 rate ends, Neon becomes the cheaper Gemini 4 model per token.
  • Antigravity in real use: whether the orchestrator setup, with Neon directing Hydrogen subagents, holds up outside Google’s own demonstrations.
  • The cyber version: how widely the Fairwind Program rollout reaches, and whether defenders see the same secure-code gains Google reports.
  • Rival pricing: whether OpenAI, Anthropic or Alibaba answer a frontier model at $3 / $15.

What is Gemini 4 Neon?

Google’s new mid-tier AI model, launched 18 November 2026. It sits between Gemini 4 Carbon, built for the hardest long-horizon projects, and Gemini 4 Hydrogen, built for fast everyday execution, and is designed for complex software engineering, knowledge work and science workflows.

How much does Gemini 4 Neon cost?

$3 per million input tokens and $15 per million output tokens, with cached input at $0.15 per million tokens, a 95% discount. At its default High thinking level, Google says it costs 32% less per finished task than Carbon.

What is the Gemini 4 Neon context window?

One million tokens. Neon can take in text, images, audio and video in a single prompt, including spreadsheets, PDFs, charts, audio and video, and turn them into structured deliverables.

How does Gemini 4 Neon score on benchmarks?

It scores 61 on the Artificial Analysis Intelligence Index at its highest thinking level, ranking third overall and just above Carbon’s 60. Google also reports first place on the Artificial Analysis Coding Agent Index at 58.6, 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at its default High setting. These are Google’s own reported figures.

What are Gemini 4 Neon’s thinking levels?

Four levels: Low, Medium, High and Extra High. High is the default. Extra High adds deeper, multi-step reasoning for the hardest engineering, analytical and scientific workloads, and records Neon’s highest intelligence score.

Joseph Zada
Written by

Joseph Zada

Joseph is the founder and publisher of ChainAIWire, where he oversees editorial direction and strategic coverage at the intersection of blockchain infrastructure, digital assets, and artificial intelligence. Driven by a mission to bridge complex emerging technologies with clear, transparent reporting, Joseph tracks the convergence of decentralized protocols, autonomous AI systems, and evolving regulatory frameworks. Under his leadership, ChainAIWire delivers objective, high quality content reporting designed for developers, institutional allocators, and industry leaders navigating the modern digital economy.

More articles by this author →

Leave a Comment