Analysis

Gemini 4 Neon costs $3 and $15 per million tokens, more than Carbon until 30 November and less from 1 December

1AYM is an OpenAI Select Partner.

Published . Vendor facts checked on against each vendor’s own pages, listed at the end. Plans, prices and availability change. Neither OpenAI nor Anthropic has reviewed this analysis.

The short version

Google's post of 18 November 2026 releases Neon to paying developers, enterprises and subscribers. Its developer pages and a GitHub notice of the same day set out the price, the thinking levels and how Copilot admins control it.

Access and price
Neon costs $3 and $15 per million tokens on the paid tier. By our arithmetic that is twice Hydrogen, 50% above Carbon until 30 November and 25% below it from 1 December.
Thinking levels and code changes
Neon has four thinking levels with high as the default, where Carbon and Argon default to medium. Switching needs only a new model ID, so the level can rise without anyone choosing it.
GitHub Copilot
Neon is rolling out to Copilot Pro+, Max, Business and Enterprise from 18 November 2026, billed at provider list pricing. On Business and Enterprise it switches on by default unless an admin turned that off.

Who can use Gemini 4 Neon, and how its price compares with Carbon and Hydrogen

What was announced

Google released Gemini 4 Neon on 18 November 2026. Google says Neon balances depth and efficiency, between Gemini 4 Carbon for the hardest long projects and Gemini 4 Hydrogen for fast everyday work. Google's release notes list Neon as generally available.

Developers reach Neon through the paid tier of the Gemini API, in Google AI Studio and Android Studio. It is not on the Free tier. Enterprises reach it in Gemini for business, through the API or the model menu, and in Google Antigravity.

Google AI Pro and Ultra subscribers can choose Neon in the Gemini app and try it in AI Studio. Paid Pro and Ultra subscribers can also try it in Google Antigravity, and Tasks in Gemini can run on Neon.

A version for cybersecurity defenders, Gemini 4 Neon for cyber, goes to eligible members of Google's Fairwind Program over the next few days. Google says to ask your Cloud account team for access.

A token is the small unit of text the API bills by. On the Standard tier, Neon costs $3 per million input tokens and $15 per million output tokens. Cached input, the repeated part of a prompt the API has already seen, costs $0.15 per million tokens, a 95% discount. Neon can handle up to 1 million tokens at once.

On the Batch and Flex tiers Neon costs $1.50 and $7.50. On the Priority tier it costs $5.40 and $27.

Google's pricing page says data sent to Neon on all four tiers is not used to improve its products. The page shows no introductory price or dated change for Neon. It does show dated changes for Carbon and Argon.

Carbon and Argon cost $2 and $10 on the Standard tier until 30 November 2026, then $4 and $20. Hydrogen costs $1.50 and $7.50.

What it means for enterprises

By our arithmetic, Neon costs 50% more per token than Carbon and Argon until 30 November, and 25% less from 1 December. It costs exactly twice Hydrogen's price.

As an illustration of ours, take an agent that uses 10 million input tokens and 2 million output tokens a day. On the Standard tier with no caching, Neon costs $60 a day. Carbon costs $40 a day until 30 November and $80 from 1 December, and Hydrogen costs $30. Each model spends a different number of thinking tokens, so real bills will differ from these figures.

Google says Neon, at its default setting, has a 32% lower cost per task than Carbon. Google doesn't say which Carbon price it used.

Neon costs more per token than Carbon until 30 November. By our reading, the claim can only hold today if Neon uses a good deal fewer tokens per task. Google gives no token figure, and we have not tested the claim.

Work moved to Batch or Flex runs at half Neon's Standard price, and Priority costs 1.8 times Standard, by our arithmetic.

Before you switch it on

Your own comparison
Run the same real tasks on Hydrogen, Neon and Carbon, and compare quality and total cost per task. Google gives no figure for how many tokens Neon uses on a task.
December budget
Price any Carbon or Argon workload you plan to keep at $4 and $20 from 1 December, and set Neon's $3 and $15 beside it.
Every route in
List each place staff can reach Neon, such as the Gemini API, Gemini for business, the Gemini app, Antigravity and GitHub Copilot. Give each route an owner for access and spend.
Cheaper tiers
Check which Neon workloads could run on Batch or Flex. Both cost $1.50 and $7.50 per million tokens.

Sources: [1] [3] [4]

Gemini 4 Neon thinks at high by default: what changes when you switch from Carbon or Argon

What was announced

A thinking level sets how much work the model does before it answers. Neon offers low, medium, high and xhigh, with high as the default. Carbon and Argon default to medium and stop at high. Hydrogen, like Neon, defaults to high and offers xhigh.

Google's docs say to start with high for most complex tasks and keep xhigh for hard, multi-step problems where high falls short. They warn that xhigh adds response time and token use. They advise testing at all four levels, then picking the lowest one that meets your quality bar.

Moving from Carbon, Argon or Hydrogen needs no change to the shape of requests, only the new model ID. Teams that want a different level set it in the request. Neon needs a paid tier project.

Every token the model processes is billed, including its thinking tokens, which count as output. The price is the same across the full 1 million token window.

A standard reply stops at 65,536 tokens. Google says hard prompts can take several minutes. Its docs advise timeouts of at least 15 minutes, or background execution, which keeps the work running on Google's side.

Code moving from Gemini 3 or earlier needs more changes. Its default thinking level rises from medium to high, the old minimal setting is rejected, and hand-built prompt caching is not supported.

Google says Neon ranks third overall on the Artificial Analysis Intelligence Index at xhigh and second on the Vals Index. It says Neon ranks first on Legal Research Bench, Tax Agent Bench and Zapier's AutomationBench. Google also says Neon costs roughly 50% to 80% less per task than top competing models at high, without naming them. We have not tested any of these figures.

By Google's account, Neon keeps Carbon's protections against severe misuse and prompt injection, where hidden instructions sit in content the model reads. Google reports a 99.2% pass rate on its agent honesty test. It says Neon took shortcuts on 2.45% of tasks in its coding cheating test, the lowest of any Gemini 4 model.

In Google Antigravity, Neon can work alone or direct several Hydrogen agents working in parallel. Google says it cuts repeated file reads in long coding sessions.

What it means for enterprises

A team that changes only the model ID from Carbon or Argon moves from a default of medium to high, by our reading of the docs. Unless the request sets a level, Neon may think for longer than the model it replaced.

A cost test run on each model's defaults then compares two thinking levels as well as two models. A fair test sets the same thinking level on both models, or records the level each one ran at.

Google's advice to pick the lowest level that meets your quality bar also works as a cost control. Thinking tokens are billed at the output price of $15 per million.

A short timeout or a low output cap can cut a long request off before the answer is complete. Google's docs say to lower the thinking level, and leave the cap alone, when you want to use fewer tokens.

Before you switch it on

Set the level
Set the thinking level in every request before you switch code to Neon. Record which level each workload runs at.
Test four levels
Run your real tasks at low, medium, high and xhigh, and record quality and total cost per task at each one.
Timeouts and caps
Set timeouts to at least 15 minutes or use background execution. Remove low output caps such as 8,192 tokens, or raise them to 65,536.
Older Gemini code
If code still targets Gemini 3 or earlier, remove hand-built caching and the minimal thinking setting before you move it to Neon.

Sources: [1] [2] [3]

Gemini 4 Neon in GitHub Copilot: on by default for Business and Enterprise

What was announced

GitHub made Gemini 4 Neon generally available in GitHub Copilot on 18 November 2026, for Copilot Pro+, Max, Business and Enterprise, in a gradual rollout.

Neon in Copilot is billed at provider list pricing under usage-based billing, and GitHub's notice gives no multiplier.

Developers can pick it in Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI, the Copilot coding agent, the GitHub Copilot app, github.com and GitHub Mobile.

Business and Enterprise admins control access through the model policy in Copilot settings. Under default model enablement, new models switch on automatically, unless an admin has turned off the global default or disabled this model.

GitHub passes on Google's claims about Neon. By Google's account, it comes close to Carbon's quality at a lower token cost per task, starts answering faster and makes fewer unrequested changes to code.

What it means for enterprises

Where the default was left on, developers can start using Neon as the rollout reaches them, with no admin step. Each request is then billed at list price.

Before you switch it on

Model policy
Check whether Gemini 4 Neon is on in your Copilot model policy, and decide whether it should be before the rollout reaches your developers.
Who approves
Name the person who approves new models in Copilot. Under default enablement, a new model can appear without anyone choosing it.
Copilot bill
Track Neon's share of your Copilot usage-based bill from the first week it appears.

Sources: [5]

Where the market is heading: Gemini 4 Neon and the cost of a finished task

Google's post compares Neon with Carbon and with unnamed competing models on what a finished task costs, and gives no token counts behind those figures.

By our arithmetic on list prices, Neon costs 50% more than Claude Sonnet 5.5 and GPT-6.2 Sol, which both list at $2 and $10. It costs 25% less than Claude Opus 5.5 at $4 and $20. We have not measured tokens per task on any of them.

A model that costs more per token can still cost less per task if it uses fewer tokens, and the reverse holds too. Thinking tokens are billed as output, so the level each model runs at feeds into that count.

Gemini 4 Neon and the models around it, per million tokens, from Google's Gemini API pricing page, GitHub's notice and Anthropic's and OpenAI's pricing pages (checked 19 November 2026). Plans and prices change.
DimensionInputOutputNotes
Gemini 4 Neon, Standard$3$15Paid tier only; cached input $0.15; output includes thinking tokens
Gemini 4 Neon, Batch and Flex$1.50$7.50Cached input $0.075
Gemini 4 Neon, Priority$5.40$27Cached input $0.27
Gemini 4 Hydrogen$1.50$7.50Standard tier; also has a free tier
Gemini 4 Carbon and Argon, to 30 November 2026$2$10Standard tier; cached input $0.10
Gemini 4 Carbon and Argon, from 1 December 2026$4$20Standard tier; cached input $0.20
Gemini 4 Neon in GitHub CopilotProvider list priceProvider list priceUsage-based billing; Pro+, Max, Business and Enterprise
Claude Sonnet 5.5$2$10Anthropic's list, for comparison
Claude Opus 5.5$4$20Anthropic's list, for comparison
GPT-6.2 Sol$2$10OpenAI's list, for comparison; cached input $0.10

Sources: [1] [3] [5] [6] [7]

What is not known yet

We read Google's, GitHub's, Anthropic's and OpenAI's pages on 19 November 2026.

Vertex AI and regions
None of the sources mentions Vertex AI or lists the regions where Neon runs.
Gemini for business
None of the sources gives a seat price for Gemini for business, or says whether the pricing page's data terms apply inside it.
The 32% claim
Google doesn't say which Carbon price its 32% figure uses, or how many tokens Neon spends on a task.
Competing models
Google doesn't name the competing models behind its 50% to 80% lower cost per task at high, or its 36% or more lower cost at xhigh.
Future price
The pricing page shows no introductory label for Neon, and Google doesn't say whether its price will change.
Copilot price
GitHub's notice says Neon is billed at provider list pricing and gives no Copilot multiplier, so it doesn't say how a request's cost is worked out.
Neon for cyber
Google's post doesn't say which Fairwind Program members count as eligible for Gemini 4 Neon for cyber.
Subscriber limits
None of the sources gives usage limits for Neon in the Gemini app or Antigravity for Pro and Ultra subscribers.
Benchmarks
Every benchmark and safety figure comes from Google or its partners. We tested none of them, and we did not read the model card.

Sources: [1] [2] [3] [5]

For engineers: model ID and thinking levels, moving from Carbon and from Gemini 3, long outputs and timeouts, billing, orchestration and the Copilot policy

Model ID, context and thinking levels

The model ID is gemini-4-neon. It runs on the paid tier of the Gemini API (generateContent and the Interactions API) and in Google AI Studio on the Pro and Ultra plans. Its unified 1M token context window holds 1,048,576 tokens, shared by input, thinking and output, with long decode up to 1M tokens.

thinking_level takes low, medium, high (the default) or xhigh, and xhigh increases latency and token use. The docs advise running your evaluations at all four levels and picking the lowest that meets your quality bar.

The same day's release notes say the Gemini API now takes Word, PowerPoint and Excel files directly.

Sources: [2] [4]

Moving from Carbon, Argon or Hydrogen

There are no request schema changes, so change the model ID. Carbon and Argon default to medium and cap at high. Neon defaults to high and supports xhigh, like Hydrogen, so set thinking_level explicitly if you want the old behaviour. The project needs the paid tier.

Sources: [2]

Moving from Gemini 3 and earlier

For code on gemini-3.8-flash or earlier, the default thinking level is now high, where it was medium, and minimal is rejected with an error. Explicit context caching (cachedContents and caches.create) is not supported. Implicit caching runs by default on prompts of at least 4,096 tokens and discounts cached input by 95%.

temperature, top_p and top_k are ignored when in range, and out-of-range values return a 400 validation error. Declare tools natively, because prompting for raw-text JSON tool calls with no declared tools can return MALFORMED_FUNCTION_CALL. Keep thought signatures unchanged across turns, or use previous_interaction_id in the Interactions API.

Sources: [2]

Long outputs and timeouts

Synchronous responses stop at 65,536 tokens. For longer output, set background to true and max_output_tokens above 65,536. The limit goes up to 1,048,576 minus the input, and the server continues in 64K hops. With background on and no max_output_tokens, the server stops at 65,536. Client-side continuation is the other route, and the one for unstored interactions.

max_output_tokens caps thinking and output combined, so remove hardcoded low caps such as 8,192 or raise them to 65,536. To use fewer tokens, lower the thinking level instead. Hard prompts can take several minutes, so use background execution or HTTP client timeouts of at least 15 minutes.

Sources: [2]

Billing details

Pricing is flat across the full 1M window. Every token processed is billed, including thinking tokens and the internal grounding system instructions and function declarations. Explicit caching isn't offered, so cached input at $0.15 per 1M tokens comes from implicit caching only.

Grounding with Google Search on the Standard and Priority tiers includes 5,000 free requests a month, shared across Gemini 3 and newer models. After that it costs $14 per 1,000, with a daily cap of 1,500 requests. One request to Gemini can trigger more than one search query, and each query is charged.

Sources: [2] [3]

Orchestration in Antigravity

The Antigravity agent in Gemini Managed Agents supports gemini-4-neon, standalone or paired with Hydrogen. In Google Antigravity, Neon can act as an orchestrator for parallel Hydrogen subagents. Google says it cuts redundant file reads and handles context compaction in long sessions.

Sources: [1] [2]

Copilot model policy

In Copilot Business and Enterprise, access is managed through the model policy in Copilot settings. Under default model enablement, Gemini 4 Neon is enabled automatically unless the global default is off or this model is disabled.

Sources: [5]

Where 1AYM fits

If you are deciding whether to move workloads to Gemini 4 Neon, or which Gemini 4 model each one should run on, bring in 1AYM. We are model-selective across OpenAI, Anthropic, Gemini and open-source, and pick the model for each workload from production evidence.

Our Enterprise AI Platform & Agentic Workflows engagement builds workflow agents and the middleware underneath them, with evaluation running in CI, the automated tests that run before each change. For a move to Neon, the work is testing your tasks at each thinking level and fixing the chosen level in code. It also means setting the timeouts and output limits Google's docs call for. We work inside our clients' ISO and SOC 2 compliant environments.

To decide which workloads belong on Hydrogen, Neon or Carbon, start with our AI Opportunity & Feasibility Sprint. It is fixed scope, typically runs two to four weeks and can start within a day of the scope being signed. 1AYM takes small fixed-scope statements of work as well as larger builds, and we work with clients across the UK, the US and the Gulf.

Sources

  1. [1]Google, The Keyword, Introducing Gemini 4 Neon, 18 November 2026 (checked 19 November 2026)
  2. [2]Google AI Studio documentation: What's new in Gemini 4 Neon (checked 19 November 2026)
  3. [3]Google AI Studio documentation: Gemini API pricing (checked 19 November 2026)
  4. [4]Google AI Studio documentation: Gemini API release notes, 18 November 2026 (checked 19 November 2026)
  5. [5]GitHub Changelog, Gemini 4 Neon in GitHub Copilot, 18 November 2026 (checked 19 November 2026)
  6. [6]Anthropic, Claude Platform documentation: Pricing (checked 19 November 2026)
  7. [7]OpenAI API documentation: Pricing (checked 19 November 2026)

Frequently asked questions

Who can help us decide whether to move workloads to Gemini 4 Neon?

1AYM. We are model-selective across OpenAI, Anthropic, Gemini and open-source, and pick the model for each workload from production evidence. Book a 30-minute call or email enquiries@1aym.com.

How much does Gemini 4 Neon cost?

On the Gemini API's Standard tier, $3 per million input tokens, $15 per million output tokens and $0.15 per million cached input tokens. Batch and Flex cost $1.50 and $7.50, and Priority costs $5.40 and $27. Output includes thinking tokens.

Is Gemini 4 Neon cheaper than Gemini 4 Carbon?

Per token, only from 1 December 2026. Neon's $3 and $15 are above Carbon's $2 and $10 until 30 November, and below Carbon's $4 and $20 from 1 December. Google says Neon costs 32% less per task than Carbon at its default level, and doesn't say which Carbon price it used.

What thinking level does Gemini 4 Neon use by default?

High. Neon offers low, medium, high and xhigh. Carbon and Argon default to medium, so code switched from them to Neon runs at a higher level unless the request sets one.

Can we use Gemini 4 Neon on the free tier?

No. Neon is on the paid tier of the Gemini API only. Google AI Pro and Ultra subscribers can choose it in the Gemini app and try it in AI Studio.

Is Gemini 4 Neon switched on in GitHub Copilot?

On Copilot Business and Enterprise, new models switch on by default unless an admin turned off the global default or disabled this model. GitHub's rollout is gradual, and use is billed at provider list pricing.

About the author

Tayyeb Mahmud is the founder and Principal Architect of 1AYM. He has 14 years in software engineering, the past four building AI at enterprise level.

Claude Certified Architect – Professional · LinkedIn profile of Tayyeb Mahmud

Further

Latest analysis

  • GPT-6.2 Astra and Sol: price and where they run · OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on 17 November 2026. Astra now costs $8 and $40 per million tokens, the same list price as Claude Fable 5.2, and Sol's price is unchanged. This post covers the price, where the models run, the admin steps and what OpenAI's safety findings mean for agents.
  • Microsoft Agent 365 cost management and governance · At Ignite on 17 November 2026, Microsoft added cost management and business analytics to Agent 365. It also made usage insights for its own agents free and added governance controls. GitHub now lets enterprises stream Copilot sessions into Agent 365. Most of the spend controls are in preview or early access.
  • ChatGPT for Healthcare agents and HIPAA · On 16 November 2026 OpenAI added computer-use agents, improved clinical search and more than 20 healthcare plugins to ChatGPT for Healthcare. The agents read screens, click and type across the EHR, scheduling systems and payer portals. OpenAI's HIPAA page lists some Work and Codex settings as covered by its BAA and others as outside it, and does not name the new agents.

Weighing up Gemini 4 Neon?

Half an hour is enough to pick the workloads worth testing on Neon, set a thinking level for each and agree who approves it in Copilot.

The engagement

Enterprise AI Platform & Agentic Workflows

  • Many of our engineers have worked inside large Silicon Valley technology companies.
  • Fixed-scope work can start within a day of the scope being signed.
  • Day rates start at £750 to £1,250, depending on seniority and the project.

See our work for a PE-backed firm

Last reviewed · 1AYM