# Google releases Gemini 4 Neon: the middle Gemini 4 model for 3 dollars

By Ivo Donker · 19 November 2026 — text: Claude (AI)
What is Gemini 4 Neon? It is the new middle model in Google's Gemini 4 line, sitting between the heavy Carbon and the fast Hydrogen. Neon targets coding, analytical reasoning and business workflows. In the Gemini API it costs 3 dollars per million input tokens and 15 dollars per million output tokens, with a 1 million token context. Google says the cost per task is 32 percent lower than with Gemini 4 Carbon.

We read the [announcement on Google's blog](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/). The launch spread on X through [a post by Logan Kilpatrick](https://x.com/OfficialLoganK/status/2123167791922741362) on November 18. The topic was also covered in the [trending briefing of November 19](https://trending.llmnet.nl/en/briefing/2026-11-19).

## What happened

Google positions Neon as the middle option of the Gemini 4 family. [Gemini 4 Carbon](https://nieuws.llmnet.nl/en/gemini-4-carbon) remains the top model for long, hard projects and [Gemini 4 Hydrogen](https://nieuws.llmnet.nl/en/gemini-4-hydrogen) the fast model for everyday work. Neon is meant to balance depth and efficiency. In Google Antigravity it can also act as an orchestrator that directs several Hydrogen subagents in parallel.

According to the blog, the model is available in the Gemini API (paid tier), in Google AI Studio and in Android Studio. Google AI Pro and Ultra subscribers can pick it in the Gemini app and in Google Antigravity, and businesses get it through Gemini for business. A separate variant for cyber defence goes to members of Google's Fairwind program over the next few days.

## Prices and benchmarks

Prices and benchmarks
Item | Gemini 4 Neon | 

Input per million tokens | 3 dollars | 

Cached input per million tokens | 0.15 dollars | 

Output per million tokens | 15 dollars | 

Context | 1 million tokens | 

Coding Agent Index (Artificial Analysis) | 58.6, first place | 

Intelligence Index (Artificial Analysis, xhigh) | 61, third place | 

Terminal-Bench 4.1 | 65.2 percent | 

Vals Index | 70.9 percent, second place | 

Neon has four thinking levels; the blog names only "high" (the default) and "xhigh". At the high level a task is said to cost 50 to 80 percent less than with competing frontier models, measured on the Vals Index and Agent Arena. On the Agent Arena Index Neon ranks second with 1336 points across more than 4,200 sessions.

Google also publishes safety figures. On Gray Swan's prompt injection test, no attack succeeded at the highest thinking level. In a test for cheating while coding, Neon took a shortcut on 2.45 percent of tasks, which Google calls the lowest rate of any Gemini 4 model. Compared with Carbon, Neon is said to under-deliver less often and to overstep its task less often.

## What this means

If you use Carbon today for ordinary coding work, Neon is the first thing to try: the context is the same size and the cost per task is lower. Cached input costs 95 percent less than regular input, which helps agents that resend the same code or documents over and over.

The competition sits in the same price range. [OpenAI's GPT-6.2 Sol](https://nieuws.llmnet.nl/en/gpt-6-2-astra-sol) costs 2 and 10 dollars per million tokens, GPT-6.2 Astra 8 and 40 dollars. Leaderboards say little about your own work: each index measures different tasks, and almost every figure comes from Google or from indexes Google itself cites. A trial on a few real tasks, with the cost per task recorded, gives the clearest picture.

## What we did not verify

We did not read Google's model card or pricing page. A model ID, an output limit and a knowledge cutoff are not in the blog. We did not re-run the benchmark scores. The claim that Neon ranks above GPT-6.2 Astra and Claude Fable 5.2 on the Coding Agent Index comes from [a post by @kimmonismus](https://x.com/kimmonismus/status/2123221275594403932) on X. The November 18 launch date comes from the post on X; the blog itself only says "today".

Checked on November 19, 2026.

This article was written with AI (Claude) and checked against the sources below.

## Sources

- Google, Introducing Gemini 4 Neon: [blog.google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/)

- Logan Kilpatrick on X about the launch: [x.com](https://x.com/OfficialLoganK/status/2123167791922741362)

- @kimmonismus on X about the Coding Agent Index: [x.com](https://x.com/kimmonismus/status/2123221275594403932)
