# Gemini 4 Neon for coding: price, benchmarks and how to use it

Authors: AutoRouter
Author profile: https://autorouter.com/about
Published: 2026-11-18T21:50:00.000Z
Updated: 2026-11-19T23:40:00.000Z
Reading time: 10 min
Canonical: https://autorouter.com/blog/gemini-4-neon
Publisher: AutoRouter, the Mac app that routes every coding agent you use to every model (https://autorouter.com)

About AutoRouter: a Mac app (macOS 14 or later) that adds every leading model to every coding agent you use (Claude Code, Codex, Cursor, etc.) and picks the model for each task, using the AI subscriptions you already pay for first. Install from https://autorouter.com/AutoRouter.dmg, or ask your agent to follow https://autorouter.com/setup.md. Routing across your plans is free; Pro is $50 a month with $20 of included usage. Unrelated to OpenRouter's Auto Router.

> Use Gemini 4 Neon in every coding agent you use with AutoRouter, the Mac app for all your coding agents: Neon is in its model picker, with no keys or endpoints to set up. Google released Neon on November 18, 2026 as the Gemini 4 model between Carbon and Hydrogen, at $3 per million input tokens and $15 per million output tokens. On Cursor's CursorBench 4.1, Neon at high scores 63.7% at $6.54 a task, against 61.5% at $8.52 for Carbon at high.

## Key facts

- As of November 19, 2026, Gemini 4 Neon costs $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens on the Gemini API's Standard tier, flat across its 1,048,576-token window ([Google](https://aistudio.google.com/docs/pricing#gemini-4-neon)).
- On CursorBench 4.1, Gemini 4 Neon at high scored 63.7% at an average of $6.54 a task, against 61.5% at $8.52 for Gemini 4 Carbon at high, 57.3% at $4.57 for Gemini 4 Hydrogen at high and 56.5% at $9.14 for Gemini 4 Argon at high ([Cursor](https://cursor.com/evals), checked November 19, 2026).
- Gemini 4 Carbon's and Argon's list prices double on December 1, 2026 to $4 input and $20 output per million tokens, while Google lists no change for Neon ([Google](https://aistudio.google.com/docs/pricing#gemini-4-carbon)).
- One assumed agent request (45,000 cached input tokens, 5,000 new input tokens and 2,000 output tokens) costs about $0.0518 on Gemini 4 Neon, against $0.0345 on Carbon or Argon through November 30, 2026 and $0.069 from December 1, at Google's list prices, by AutoRouter's arithmetic.
- Google's November 18, 2026 docs give Gemini 4 Neon complex software engineering, keep Carbon for the hardest architecture, debugging, science and security problems, and keep Hydrogen as the default for everyday coding ([Google](https://aistudio.google.com/docs/whats-new-gemini-4-neon#choosing-between-models)).

![AutoRouter puts Neon in every coding agent. 1, Enable each agent, in AutoRouter's settings. 2, Open the model picker, model menu or /model. 3, Choose Gemini 4 Neon, or name it in a rule. 4, Neon runs the task, model and cost shown. Source: AutoRouter setup guide; Gemini 4 Neon released November 18, 2026.](https://autorouter.com/blog/images/gemini-4-neon/1-use-neon.png)

Gemini 4 Neon is in the model picker of [AutoRouter](https://autorouter.com), the Mac app for all your coding agents, so you can choose it in every coding agent you use (Claude Code, Codex, Cursor, etc.) with no Google API key or endpoint to set up. Google [released Neon on November 18, 2026](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/) as the Gemini 4 model that balances depth and efficiency, between Carbon for the hardest long-horizon work and Hydrogen for fast everyday execution. It [costs](https://aistudio.google.com/docs/pricing#gemini-4-neon) $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens, as of November 19, 2026. On Cursor's [CursorBench 4.1](https://cursor.com/evals), Neon at its default high setting scored 63.7% at an average of $6.54 a task, ahead of Gemini 4 Carbon at high (61.5%, $8.52 a task), Gemini 4 Hydrogen at high (57.3%, $4.57) and Gemini 4 Argon at high (56.5%, $9.14).

AutoRouter wrote this post. Every outside figure below comes from Google's, Cursor's and OpenRouter's own pages, linked where it is used and checked on November 19, 2026.

## What is Gemini 4 Neon?

Gemini 4 Neon is Google's Gemini 4 model for complex coding, analytical reasoning and production work, placed between Hydrogen and Carbon. Google [announced it on November 18, 2026](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/), and its [model page](https://aistudio.google.com/docs/models/gemini-4-neon) and [What's new in Gemini 4 Neon](https://aistudio.google.com/docs/whats-new-gemini-4-neon) give these details as of November 19, 2026:

- **Where it runs:** the Gemini API as `gemini-4-neon`, on paid-tier projects only, in Google AI Studio and Android Studio, Gemini for business, Google Antigravity for paid Google AI Pro and Ultra subscribers, and the Gemini app.
- **Size:** one window of 1,048,576 tokens shared by input and output. A synchronous response stops at 65,536 tokens unless you use background mode for long output.
- **Input:** text, image, video, audio and PDF. It outputs text only.
- **Thinking:** `low`, `medium`, `high` or `xhigh`, with `high` the default. Thinking tokens bill as output tokens.
- **Tools:** function calling, structured outputs, code execution, computer use, Google Search grounding and URL context. Caching is implicit only; explicit caching isn't supported.

## Where does Neon sit against Carbon, Hydrogen and Argon?

Neon sits between Hydrogen and Carbon. Google's [Neon docs](https://aistudio.google.com/docs/whats-new-gemini-4-neon#choosing-between-models) give Neon complex software engineering, knowledge work and science, keep Carbon for the hardest architecture, debugging, science and security problems, and keep Hydrogen as the default for everyday coding and agentic work. They give no advice on Neon against Argon; Google's [Hydrogen docs](https://aistudio.google.com/docs/whats-new-gemini-4-hydrogen#choosing-between-models) point advanced long-horizon tasks to Argon.

| Gemini 4 model, November 19, 2026 | Google points it to                                                | Thinking levels                       | Input / output per million tokens                                           | From December 1, 2026 |
| --------------------------------- | ------------------------------------------------------------------ | ------------------------------------- | --------------------------------------------------------------------------- | --------------------- |
| Carbon                            | The hardest architecture, debugging, science and security problems | Up to `high`; `medium` is the default | [$2 / $10](https://aistudio.google.com/docs/pricing#gemini-4-carbon)        | $4 / $20              |
| **Neon**                          | Complex software engineering, knowledge work and science           | Up to `xhigh`; `high` is the default  | [$3 / $15](https://aistudio.google.com/docs/pricing#gemini-4-neon)          | No change listed      |
| Hydrogen                          | Everyday coding and agentic work                                   | Up to `xhigh`; `high` is the default  | [$1.50 / $7.50](https://aistudio.google.com/docs/pricing#gemini-4-hydrogen) | No change listed      |
| Argon                             | Advanced long-horizon tasks, in Google's Hydrogen docs             | Up to `high`; `medium` is the default | [$2 / $10](https://aistudio.google.com/docs/pricing#gemini-4-argon)         | $4 / $20              |

Moving to Neon from another Gemini 4 model needs only the new model ID, but check the thinking level: Neon starts at `high`, while Carbon and Argon start at `medium`. [Gemini 4 Carbon vs Gemini 4 Argon](/blog/gemini-4-carbon-vs-argon) and [Gemini 4 Hydrogen vs Gemini 4 Argon](/blog/gemini-4-hydrogen-vs-argon) cover the rest of the family, and [which model to use for which coding task](/blog/which-model-for-which-task) maps the rest of your work to a model tier.

## What does Gemini 4 Neon cost per request?

Gemini 4 Neon costs 50% more per token than Carbon and Argon through November 30, 2026, and 25% less from December 1, when Google's [price list](https://aistudio.google.com/docs/pricing#gemini-4-neon) doubles theirs. Hydrogen costs half of Neon's price. Google lists no cache-write price for Neon, so new input bills at the input price.

![From December 1, Neon undercuts Carbon per token. Output price per million tokens: Gemini 4 Carbon, $10, doubles to $20 on December 1; Gemini 4 Argon, $10, doubles to $20 on December 1; Gemini 4 Neon, $15, no change listed; Gemini 4 Hydrogen, $7.50, no change listed. Source: Gemini Developer API pricing, checked November 19, 2026.](https://autorouter.com/blog/images/gemini-4-neon/3-output-price.png)

The arithmetic below prices the assumed agent request this blog's pricing posts use: 45,000 cached input tokens, 5,000 new input tokens and 2,000 output tokens, thinking included. It assumes all 45,000 repeated tokens hit Google's implicit cache, which Google does not promise.

```text
Assumed request: 45,000 cached input, 5,000 new input, 2,000 output tokens
Google's Standard list prices per million tokens as of November 19, 2026

Gemini 4 Neon
  45,000 cached input  x  $0.15  =  $0.00675
   5,000 new input     x  $3     =  $0.015
   2,000 output        x  $15    =  $0.03
  Total                             $0.05175, about $0.0518

Gemini 4 Hydrogen                               $0.0259
Gemini 4 Carbon or Argon, through November 30   $0.0345
Gemini 4 Carbon or Argon, from December 1       $0.0690
```

The arithmetic gives every model the same token counts, and real counts differ. Neon thinks at `high` unless you set a lower level, while Carbon starts at `medium`, and Google claims Neon still costs less per task than Carbon. A per-task measurement shows how that balances out.

## How does Neon score on CursorBench, and what does a task cost?

![Neon at high beats Carbon for less per task. CursorBench 4.1 score: Gemini 4 Neon, high, 63.7%, $6.54 a task; Gemini 4 Carbon, high, 61.5%, $8.52 a task; Gemini 4 Hydrogen, high, 57.3%, $4.57 a task; Gemini 4 Argon, high, 56.5%, $9.14 a task. Source: Cursor, CursorBench 4.1 score and average cost per task, checked November 19, 2026.](https://autorouter.com/blog/images/gemini-4-neon/2-cursorbench.png)

On [CursorBench 4.1](https://cursor.com/evals), Cursor's benchmark of ambiguous, multi-file tasks from real Cursor sessions, Gemini 4 Neon at high scored 63.7% at an average of $6.54 a task, as of November 19, 2026. Only Neon at Cursor's Extra High setting scored higher, at 65.4% and $10.67 a task. Cursor computes cost per task by applying each model's published per-million-token prices for input, cache reads, cache writes and output to the tokens it used on each task.

| Model and setting, as Cursor labels it | CursorBench 4.1 score | Average cost per task | Tokens per task | Steps per task |
| -------------------------------------- | --------------------- | --------------------- | --------------- | -------------- |
| Gemini 4 Neon Extra High               | 65.4%                 | $10.67                | 136,273         | 73             |
| **Gemini 4 Neon High**                 | **63.7%**             | **$6.54**             | 88,596          | 56             |
| Gemini 4 Carbon High                   | 61.5%                 | $8.52                 | 114,518         | 97             |
| Gemini 4 Hydrogen Extra High           | 61.4%                 | $8.09                 | 205,981         | 113            |
| Claude Fable 5.2 Max                   | 61.3%                 | $48.11                | 541,003         | 295            |
| Gemini 4 Neon Medium                   | 57.7%                 | $3.07                 | 47,059          | 41             |
| Gemini 4 Hydrogen High                 | 57.3%                 | $4.57                 | 113,807         | 77             |
| Claude Opus 5.5 Extra High             | 56.7%                 | $6.54                 | 104,901         | 95             |
| Gemini 4 Argon High                    | 56.5%                 | $9.14                 | 135,038         | 118            |

Four things to keep in mind when you read these numbers:

- **Cursor measured it, in Cursor's own agent.** It is one benchmark, and Cursor notes that small differences in scores may not be statistically meaningful.
- **Neon's lower cost comes from doing less.** At high, Neon used 88,596 tokens and 56 steps per task against Carbon's 114,518 tokens and 97 steps, so it cost 23% less per task even at a higher price per token.
- **Carbon's and Argon's costs will double.** Every Carbon and Argon rate doubles on December 1, 2026, so the same tokens will cost twice as much. Neon's price list shows no change.
- **CursorBench 4.1 is new.** Cursor's changelog dates it November 18, 2026, with changed instruction-following problems and a longer agent time limit, so its scores don't line up with CursorBench 4.0 numbers quoted elsewhere.

## What does Google claim about Gemini 4 Neon, and who measured it?

Google claims Neon tops Artificial Analysis's coding-agent index and costs less per task than Carbon. Every figure below is Google's, from its [November 18, 2026 announcement](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/), with the thinking level where Google names one.

| Claim, as Google states it on November 18, 2026                    | Thinking level Google names | Who ran it                              |
| ------------------------------------------------------------------ | --------------------------- | --------------------------------------- |
| Artificial Analysis Intelligence Index: 61, third overall          | `xhigh`                     | Artificial Analysis, reported by Google |
| Artificial Analysis Coding Agent Index: 58.6, first overall        | Not named                   | Artificial Analysis, reported by Google |
| Terminal-Bench 4.1: 65.2%                                          | Not named                   | Google                                  |
| FrontierCode 2.0: 46.2%                                            | `high`, the default         | Reported by Google, linking Cognition   |
| 32% lower cost per task than Carbon on Artificial Analysis's index | `high`, the default         | Google                                  |

Google's 32% is measured on a broad index that also covers knowledge work and science. On CursorBench 4.1, a coding-only benchmark run by Cursor, the gap at high is 23%. To judge your own work, run one everyday task and one hard task from your backlog on Neon and on the model you use now, and compare the results and the token counts.

## How do you use Gemini 4 Neon in your coding agent?

Pick Gemini 4 Neon in your coding agent's model picker with AutoRouter, or use Google's own agent or a gateway that lists the model. As of November 19, 2026, Google's docs have no guide for running Neon inside third-party coding agents.

- **With AutoRouter.** Enable each coding agent in AutoRouter, then choose Gemini 4 Neon in the picker you already use: the model menu in Codex, `/model` in Claude Code, etc. No Google project or key is needed.
- **In Google Antigravity.** Google's [announcement](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/) puts Neon in Antigravity for paid Google AI Pro and Ultra subscribers.
- **In Codex, with your own key.** A custom provider has to speak OpenAI's Responses API, the only `wire_api` value Codex supports ([config reference](https://learn.chatgpt.com/docs/config-file/config-reference)), and Google's [OpenAI compatibility page](https://aistudio.google.com/docs/openai) documents a Chat Completions endpoint, so requests go through a gateway. [OpenRouter lists the model](https://openrouter.ai/google/gemini-4-neon) as `google/gemini-4-neon`, served by Google Vertex and Google AI Studio at $3 input and $15 output, and publishes a [Codex guide](https://openrouter.ai/docs/cookbook/coding-agents/codex-cli).
- **In Claude Code, with your own key.** Claude Code connects to a gateway through one of the [API formats its docs list](https://code.claude.com/docs/en/llm-gateway-protocol#api-formats), and Google's Gemini API docs describe none of them, so requests pass through a gateway that translates. OpenRouter publishes a [Claude Code guide](https://openrouter.ai/docs/cookbook/coding-agents/claude-code-integration).

AutoRouter has not run Neon through OpenRouter for this post, so follow OpenRouter's guides for the exact settings, and remember that the Gemini API needs a paid-tier project for Neon. [Try a new model in Codex or Claude Code without another API key](/blog/try-new-models-in-codex-and-claude-code) covers the gateway routes.

## How does AutoRouter handle Gemini 4 Neon?

Gemini 4 Neon is in AutoRouter's model picker, where you can choose it yourself or name it in a plain-English rule. AutoRouter, the Mac app for all your coding agents, [puts new models in the picker you already use](https://autorouter.com/#product) right away, with no keys or endpoints to set up, and lists every model with its role and price at [autorouter.com/models](https://autorouter.com/models).

- **Choose it.** Pick Gemini 4 Neon in your coding agent's model picker, and your prompts go straight to it.
- **Name it in a rule.** Rules are plain English, for example "Use Gemini 4 Neon for refactors across several files." Your rules come before AutoRouter's own.
- **Or select AutoRouter.** It works out how hard each task is and only considers models that clear the quality bar for it. Hard work goes to the most capable model your plans cover, and to credits only when none is strong enough.
- **See what happened.** Every request shows its model, what it cost and why.

AutoRouter is free to start, and routing across the AI subscriptions you already pay for is free. Models outside your plans, Neon included, come with AutoRouter Pro and its monthly included usage, and paid models bill at the provider's price plus 10%, which AutoRouter earns. The AutoRouter team tests every new model as it becomes available. AutoRouter runs on macOS 14 or later and is in beta, and requests pass through its servers on their way to each provider ([AutoRouter privacy](https://autorouter.com/privacy)).

## Questions

### Is Gemini 4 Neon better than Gemini 4 Carbon for coding?

On CursorBench 4.1 at high, Neon scored slightly higher and cost less: 63.7% at $6.54 a task against Carbon's 61.5% at $8.52, as of November 19, 2026 ([Cursor](https://cursor.com/evals)). Google's docs still keep the hardest architecture, debugging, science and security problems on Carbon. Test both on a hard task from your own backlog.

### Is Gemini 4 Neon cheaper than Gemini 4 Carbon?

Per token, not until December 1, 2026: Neon costs $3 input and $15 output per million tokens against Carbon's $2 and $10, and Carbon rises to $4 and $20 on December 1 ([Google](https://aistudio.google.com/docs/pricing#gemini-4-neon)). Per task, Cursor already measured Neon at high as 23% cheaper than Carbon at high, because Neon used fewer tokens and steps.

### Which thinking level should you use with Gemini 4 Neon?

Start at `high`, the default, which Google recommends for most tasks. Google's [docs](https://aistudio.google.com/docs/whats-new-gemini-4-neon#understanding-thinking-levels) suggest `medium` for interactive coding and live UI prototyping, `low` for cost- or latency-sensitive work, and `xhigh` only for hard multi-step problems where `high` falls short. On CursorBench 4.1, Neon at medium scored 57.7% at $3.07 a task, close to Hydrogen at high.

### Can you use Gemini 4 Neon on the Gemini API's free tier?

No. Google's [migration checklist](https://aistudio.google.com/docs/whats-new-gemini-4-neon#migration-checklist) says Neon needs a paid-tier project, and its prices apply to the paid tier only. Through AutoRouter, Neon bills from AutoRouter Pro's included usage and credits instead of a Google project.

## Sources

- [Introducing Gemini 4 Neon](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/), Google, November 18, 2026
- [Gemini 4 Neon](https://aistudio.google.com/docs/models/gemini-4-neon), Gemini API docs, Google
- [What's new in Gemini 4 Neon](https://aistudio.google.com/docs/whats-new-gemini-4-neon), Gemini API docs, Google
- [What's new in Gemini 4 Hydrogen](https://aistudio.google.com/docs/whats-new-gemini-4-hydrogen), Gemini API docs, Google
- [Gemini Developer API pricing](https://aistudio.google.com/docs/pricing), Google, checked November 19, 2026
- [OpenAI compatibility](https://aistudio.google.com/docs/openai), Gemini API docs, Google
- [CursorBench 4.1](https://cursor.com/evals), Cursor, checked November 19, 2026
- [Gemini 4 Neon on OpenRouter](https://openrouter.ai/google/gemini-4-neon), checked November 19, 2026
- [Configuration reference](https://learn.chatgpt.com/docs/config-file/config-reference), Codex docs, OpenAI
- [LLM gateway API formats](https://code.claude.com/docs/en/llm-gateway-protocol#api-formats), Claude Code docs, Anthropic
- OpenRouter guides for [Codex](https://openrouter.ai/docs/cookbook/coding-agents/codex-cli) and [Claude Code](https://openrouter.ai/docs/cookbook/coding-agents/claude-code-integration)

## Read next

- [Gemini 4 Hydrogen vs Gemini 4 Argon: which to use for coding](https://autorouter.com/blog/gemini-4-hydrogen-vs-argon.md): Use both with AutoRouter, the Mac app for all your coding agents: Gemini 4 Hydrogen and Gemini 4 Argon are both in its model picker, with no keys or endpoints to set up. Picking by hand, move everyday coding, code edits, test runs and subagent work to Gemini 4 Hydrogen, released November 12, 2026, and keep Gemini 4 Argon or Gemini 4 Carbon for long, hard reasoning. Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, against Argon's $2 and $10 through November 30 and $4 and $20 from December 1.
- [Gemini 4 Carbon vs Gemini 4 Argon: which to use for coding](https://autorouter.com/blog/gemini-4-carbon-vs-argon.md): Use both with AutoRouter, the Mac app for all your coding agents: Gemini 4 Carbon and Gemini 4 Argon are both in its model picker, with no keys or endpoints to set up. Picking by hand, keep everyday coding on Gemini 4 Argon, and bring in Gemini 4 Carbon, generally available since November 10, 2026, for root-cause debugging in large codebases, architecture changes, migrations and security audits. The two models share one price list: $2 per million input tokens and $10 per million output tokens through November 30, then $4 and $20, and Google says Carbon uses more tokens on each request.
- [Muse Ember vs Claude Haiku 5.5: Meta's low-cost model for coding](https://autorouter.com/blog/muse-ember-vs-claude-haiku-5-5.md): Use both with AutoRouter, the Mac app for all your coding agents: Muse Ember and Claude Haiku 5.5 are both in its model picker, with no keys or endpoints to set up. Muse Ember is Meta's smallest Muse model, released on November 9, 2026 at $0.10 per million input tokens and $0.50 per million output tokens, the same list prices as Claude Haiku 5.5 and GPT-6 Luna on prompts up to 100,000 tokens. Meta says it matches Haiku 5.5's best score on the Artificial Analysis Intelligence Index at about a third of the cost per task, and Meta's own docs keep multi-file changes on a larger model.

Tags: [Models](https://autorouter.com/blog/tags/models.md), [Comparisons](https://autorouter.com/blog/tags/comparisons.md), [Costs](https://autorouter.com/blog/tags/costs.md)

Drafted with AI from the sources linked above, then checked and edited by the AutoRouter team.
