Google Gemini 4 Neon Launches: Faster AI, Lower Cost

Published (Updated )

The Quick Answer
- Google launched Gemini 4 Neon on November 18, 2026, the third model in the Gemini 4 family.
- Neon sits between Hydrogen, the fast model for everyday tasks, and Carbon, the flagship for the hardest long-running work.
- Google says Neon delivers "frontier-level intelligence" with fast end-to-end task completion across coding, knowledge work and science.
- It scores 61 on the Artificial Analysis Intelligence Index at its highest thinking level, ranking third overall β ahead of Claude Opus 5.5 (59) and GPT-6.2 Sol (57), according to Google.
- It ranks first overall on the Artificial Analysis Coding Agent Index, with a score of 58.6.
- Pricing is $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window.
- It is available now in the Gemini API, Google AI Studio, the Gemini app for paid Pro and Ultra subscribers, and in Google Antigravity.
- The benchmark numbers below are Google's own reported figures, and independent leaderboards should be checked as they update.
What Gemini 4 Neon Is
Google DeepMind introduced Gemini 4 in stages. Carbon and Hydrogen arrived first, and Neon followed on November 18 as the middle model of the family.
- Hydrogen handles fast, everyday execution.
- Carbon tackles the hardest, long-horizon projects.
- Neon balances depth and efficiency: complex work that needs advanced reasoning, done at high frequency and lower cost.
Google's own description is that Neon brings "frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows."
The model supports text, image, audio and video input, and has a context window of up to 1 million tokens for input and output. That means it can take in very large documents, codebases, spreadsheets and research material in a single session and turn them into structured deliverables.
Neon also comes with four thinking levels β low, medium, high and xhigh β so users can dial the reasoning effort up or down depending on the task. The default setting is high. Higher thinking levels cost more tokens but produce stronger results on hard problems.
The Benchmark Numbers
Google's launch post is packed with benchmark claims. Here are the headline figures, all as reported by Google.
Overall intelligence
- Artificial Analysis Intelligence Index: 61 at the xhigh thinking level, ranked third overall, just behind larger and slower models.
- Vals Index: ranked second overall with 70.9 percent, landing on the Pareto frontier of capability versus cost.
- Agent Arena Index: debuted second globally with a score of 1336 across more than 4,200 real-world agent sessions.
- Steerability: ranked second on the Agent Arena Index, measuring how cleanly the model adapts when a user corrects it mid-task.
Coding
- Artificial Analysis Coding Agent Index: ranked first overall with a score of 58.6.
- Terminal-Bench 4.1: 65.2 percent, testing the ability to carry out tasks in a terminal.
- FrontierCode 2.0: 46.2 percent at the default high thinking setting, testing advanced software development.
- CursorBench 4.1: 65.4 percent, testing performance inside AI coding tools.
Knowledge work and business
- Legal Research Bench: ranked first.
- Tax Agent Bench: ranked first.
- Surge AI benchmarks: ranked first on most tests across finance, accounting, spreadsheet modelling and complex constraint following.
- Chartography: 73.5 percent, for reading dense professional charts.
- Agents' Last Exam v1.1: 75.8 percent, on par with leading models.
- Zapier AutomationBench: ranked first overall with 56.3 percent.
- Andon Labs Vending-Bench 2: ranked first with $16,843, a long-horizon simulation of running a business.
Science and maths
- Terminal-Bench Science: best in class at 70.5 percent.
- RiemannBench: best in class at 84.0 percent.
- HLE-Diamond: best in class at 71.3 percent.
Google also says Neon records its lowest fabrication rate yet in internal Google Antigravity testing β in other words, it makes things up less often than earlier models when tested inside Google's own coding environment.
One important note: most of these numbers come from Google's launch post, and several of the named benchmarks are run by outside groups such as Artificial Analysis, Vals, Arena, Zapier and Andon Labs. Google does not publish raw test logs in the post, so the ranks should be treated as vendor-reported until they appear on the third-party leaderboards themselves.

What It Costs
Neon's pricing is straightforward, and it is the model's main selling point.
- Input tokens: $3 per million.
- Output tokens: $15 per million.
- Cached input tokens: $0.15 per million, a 95 percent discount on the input price.
- Context window: 1 million tokens.
Google says that at the default high thinking level, Neon costs 32 percent less per task than Carbon, and roughly 50 to 80 percent less per task than top competing frontier models on the Vals and Agent Arena Pareto charts. Those comparisons are Google's own calculations.
There is one pricing quirk worth knowing about.
- Carbon's list price is $4 per million input tokens and $20 per million output tokens.
- But Carbon is currently on a promotional price of $2 input and $10 output, running through November 30.
- Until that promotion ends, Carbon is actually cheaper per token than Neon.
- After November 30, the order reverses: Hydrogen is cheapest, then Neon, then Carbon.
For anyone choosing between the models, the per-token price is only part of the story. The cost per completed task matters more, and that depends on how many tokens a model needs to finish the job. Google's claim is that Neon finishes complex work with fewer wasted tokens than its rivals, which is why its cost per task is lower even when its per-token price is not the cheapest.

Where You Can Use It
Neon is rolling out across Google's AI products.
- Gemini API in Google AI Studio: available to developers now.
- Gemini for business: available through the API and the model dropdown.
- Gemini app: available to paid Google AI Pro and Ultra subscribers.
- Google Antigravity: available to paid AI Pro and Ultra subscribers.
The launch also reshuffles the model picker for paid users.
- Neon replaces Argon, the previous mid-tier model.
- Gemini 3.1 Pro is being retired.
- Gemini 3.5 Flash-Lite is now labelled "Auto" in the model picker for paid users.
For most Gemini app users on a paid plan, Neon becomes the default choice for serious work: stronger than the fast everyday models, cheaper and quicker than Carbon.
Who Neon Is For
Google is pitching Neon at three groups in particular.
Developers and coding teams
Coding is where Neon posts its strongest results. A first-place ranking on the Coding Agent Index, strong Terminal-Bench and FrontierCode scores, and the ability to adjust thinking levels per task make it a natural fit for teams that use AI to write, review and debug code all day.
- Complex software engineering tasks that need sustained reasoning.
- Terminal and command-line work, where the model has to run tools and check its own output.
- Long coding sessions where cost per task matters because the model is called hundreds of times a day.
Businesses and analysts
Neon's knowledge-work results point at office and enterprise use.
- Finance, accounting and spreadsheet modelling, where it ranks first on several Surge AI tests.
- Legal research and tax work, where it ranks first on both specialist benches.
- Reading dense charts and documents, and turning them into structured reports.
- Automating multi-step workflows, where its AutomationBench and Vending-Bench results suggest it can run a process from start to finish.
Researchers and scientists
The science benchmarks are the third pillar of the launch.
- Best-in-class results on Terminal-Bench Science, RiemannBench and HLE-Diamond.
- A 1 million-token context window that can hold large research papers, datasets and codebases together.
- The ability to synthesise across text, images, audio and video in a single session.
For all three groups, the pitch is the same: close to flagship quality, at a mid-tier price, and fast enough to use all day.
Neon vs Carbon vs Hydrogen
The Gemini 4 family now has three models, and choosing between them comes down to the trade-off between capability, speed and cost.
- Hydrogen is the fastest and cheapest. It is built for everyday tasks where speed matters more than depth.
- Neon is the middle model. It handles complex coding, analysis and science work, with frontier-level results on many benchmarks, at a lower cost per task than Carbon.
- Carbon is the flagship. It is the most capable model in the family, designed for the hardest and longest-running projects, where maximum reasoning matters more than speed or cost.
Google's own framing is that Neon offers Carbon quality at Flash speeds. The benchmark numbers support that claim in coding and knowledge work, where Neon matches or beats Carbon on several tests. Carbon still leads on the longest and hardest reasoning tasks, where its larger scale pays off.
The four thinking levels add another dimension. A team can run Neon on low for routine work and switch to xhigh for the hardest problems, without changing models. That flexibility is part of why Google expects Neon to become the workhorse of the family.
The Bigger Picture
Neon is Google's fourth Gemini 4 model release in about seven weeks, and the pace is part of the story.
- The AI model race in 2026 is being fought on benchmarks, price and speed at the same time.
- Every major lab is releasing new models every few weeks, and each one claims to lead on some combination of capability and cost.
- Google's strategy with Gemini 4 is to cover the whole range: a fast cheap model, a mid-tier workhorse and a flagship, all sharing the same 1 million-token context window and the same thinking-level controls.
- Neon is the model most people will actually use day to day, which makes it the most important of the three for Google's paying customers.
There is also a competitive angle. Neon's Intelligence Index score of 61 puts it ahead of Claude Opus 5.5 at 59 and GPT-6.2 Sol at 57, according to Google's figures. If those numbers hold up on independent leaderboards, Neon becomes the strongest mid-tier model available β and a direct challenge to rivals that charge flagship prices for similar capability.
The next test will be what users actually experience. Benchmarks measure models in controlled conditions. Real coding sessions, messy spreadsheets and long research documents are harder. Neon's steerability score β how well it adapts when a user corrects it mid-task β may matter more in practice than any single benchmark, because that is what a long working session with an AI actually feels like.
For now, the facts are clear: Neon is live, it is priced at $3 per million input tokens and $15 per million output tokens, it leads the coding agent rankings, and it is available to Pro and Ultra subscribers in the Gemini app and Antigravity, and to developers through the Gemini API.
Key Points at a Glance
- Google launched Gemini 4 Neon on November 18, 2026, as the middle model of the Gemini 4 family.
- It sits between Hydrogen (fast, everyday) and Carbon (flagship, hardest work).
- It scores 61 on the Artificial Analysis Intelligence Index at xhigh, ranking third overall β ahead of Claude Opus 5.5 and GPT-6.2 Sol, per Google.
- It ranks first overall on the Artificial Analysis Coding Agent Index (58.6) and first on the Legal Research, Tax Agent, AutomationBench and Vending-Bench 2 tests.
- Pricing is $3 per million input tokens, $15 per million output tokens and $0.15 per million cached input tokens, with a 1 million-token context window.
- It has four thinking levels (low, medium, high, xhigh); the default is high.
- Google reports 32 percent lower cost per task than Carbon at the default level, and 50β80 percent lower than top competing frontier models.
- It is available in the Gemini API, Google AI Studio, the Gemini app for Pro and Ultra subscribers, and Google Antigravity.
- Neon replaces Argon, and Gemini 3.1 Pro is being retired; Carbon's promotional $2/$10 pricing runs through November 30.
FAQs
What is Gemini 4 Neon?
Gemini 4 Neon is a mid-tier AI model launched by Google on November 18, 2026. It sits between Hydrogen, the fast everyday model, and Carbon, the flagship, and is designed for complex coding, knowledge work and science at a lower cost.
How much does Gemini 4 Neon cost?
Neon costs $3 per million input tokens and $15 per million output tokens, with cached input tokens at $0.15 per million. It has a 1 million-token context window. Google reports 32 percent lower cost per task than Carbon at the default high thinking level.
How does Gemini 4 Neon rank on benchmarks?
Google reports Neon scores 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level (third overall), ranks first on the Coding Agent Index (58.6), and first on the Legal Research Bench, Tax Agent Bench, AutomationBench and Vending-Bench 2. These are Google-reported figures.
Where can I use Gemini 4 Neon?
Neon is available in the Gemini API in Google AI Studio, in Gemini for business, in the Gemini app for paid Google AI Pro and Ultra subscribers, and in Google Antigravity for Pro and Ultra subscribers.
What are the thinking levels in Gemini 4 Neon?
Neon has four thinking levels β low, medium, high and xhigh β with high as the default. Higher levels use more tokens but give stronger results on hard problems, so users can adjust reasoning effort per task.
Is Gemini 4 Neon better than Gemini 4 Carbon?
Neon matches or beats Carbon on several coding and knowledge-work benchmarks at a lower cost per task, but Carbon remains the flagship for the hardest, longest-running reasoning tasks. Carbon is also on a promotional $2/$10 per million token price through November 30, 2026.
Was this article helpful?

Written By
TechorCoffeeContributing Writer
TechorCoffee is a technology reporter and software creator writing reviews and tutorials for TechOrCoffee.
Comments (0)
Leave a Comment
Read Next

Tue Oct 13 2026
Gemini 4 Argon: Everything You Need to Know
Gemini 4 Argon is Googleβs new flagship AI model. See its benchmarks vs GPT-6 Astra and Claude, pricing, 1M-token output, safety features and how to get access.

Tue Oct 13 2026
Google Gemini 4 Argon: Features, Benchmarks, Pricing & Guide
Discover Google DeepMind's Gemini 4 Argon. Explore its 1M token output limit, #1 Vals Index ranking, API pricing, coding capabilities, and enterprise use cases.

Fri Oct 30 2026
DeepSeek V4.1 Flash 1017: Cheap Coding AI Explained
DeepSeek V4.1 Flash 1017 explained for developers: 77.0 DeepSWE score, faster shorter outputs, Search and Fetch APIs, MIT open weights, and off-peak pricing.