SuperQode
Gemini 4 Neon
CodeGem
Google

Gemini 4 Neon and the rest of Google's lineup in SuperQode

Gemini 4 Neon is in SuperQode's Google provider, in the Mac model menu, and in CodeGem, SuperQode's native harness for Gemini. Neon is Google's heavy Gemini 4 model for complex coding and long agentic sessions. It is available on the paid tier of the Gemini API only.

November 20, 2026
6 min read
By Superagentic AI
Gemini 4 Neon and the rest of Google's lineup in SuperQode

SuperQode β€’ Gemini 4 Neon β€’ CodeGem

Gemini 4 Neon and the rest of Google's lineup in SuperQode

The model id is gemini-4-neon. It has a 1,048,576 token context window and a 65,536 token output limit. Thinking levels are low, medium, high (the default) and xhigh. minimal is not supported. Paid-tier prices, per 1 million tokens, are $3.00 input, $0.15 cached input and $15.00 output. Thinking tokens are billed as output.

SuperQode keeps the tools, approval prompts and edits on your machine. Google sees the API requests from your key. Create a key in Google AI Studio, then connect with GEMINI_API_KEY or GOOGLE_API_KEY. Either name works.

Watch the Demo

Watch on YouTube

The Gemini Lineup in SuperQode

These are the reviewed Gemini chat models SuperQode lists first. They all share the 1,048,576 token context window. The thinking defaults below are the ones SuperQode documents and, for Gemini 4, the ones CodeGem sends when you have not set a level yourself.

Model Best for Thinking Price per 1M (input / cached / output)
gemini-4-neonComplex coding and agentic work (paid tier)low, medium, high (default), xhigh$3.00 / $0.15 / $15.00
gemini-4-hydrogenEveryday Gemini 4 codingsame levels as Neon, default high$1.50 / $0.075 / $7.50
gemini-4-argonHard, long-horizon coding (account access)low, medium (default), high$2.00 / $0.10 / $10.00
gemini-3.8-flashEveryday Flash coding and agent worklow, medium (default), high$0.75 / $0.075 / $3.75
gemini-3.6-flashStable Flash alternativeminimal, low, medium (default), high$0.75 / $0.075 / $3.75
gemini-3.5-flash-liteSubagents and quick tasksminimal (default), low, medium, high$0.30 / $0.03 / $2.50
gemini-3.1-flash-liteHigh-volume, lower-cost workleave at Google's default$0.25 / $0.025 / $1.50
gemini-3.1-pro-previewLarge codebases and careful reviewslow, medium, high (default)$2.00 / $0.20 / $12.00

Argon's introductory prices rise on 1 December 2026 to $4.00 / $0.20 / $20.00. The 3.8 and 3.6 Flash prices rise on 1 January 2027 to $1.50 / $0.15 / $7.50. Gemini 3.1 Pro prompts over 200,000 tokens use $4.00 / $0.40 / $18.00. There is also gemini-3.1-pro-preview-customtools, a separate preview endpoint with the same limits and prices, for work that should lean on custom tools. The Google list also includes Gemma 4 (gemma-4-31b-it and gemma-4-26b-a4b-it) on the Gemini API free tier, which is a different family.

New Gemini Models Appear Without an App Update

With the latest SuperQode, the Google menu can add new Gemini chat models from the cached models.dev catalog. They appear after the reviewed list, newest first, at most four at a time. A model is added only when it supports tool calling, its id is a Gemini 3 or newer chat model, and it was released no more than 60 days before the newest reviewed model, or later. Rolling -latest aliases, retired ids, and embedding, image, speech, live, video and research models are left out. The reviewed ids stay first, so the default choice does not jump to a new launch.

If the catalog cache is missing, or the Mac app is offline, you get the reviewed list alone. The Mac app says "Showing the saved catalog" or "Showing built-in model examples." In the terminal, run :models update, or press R in the :connect byok picker, to refresh the cache.

Auto-added models run with generic settings until SuperQode tunes them. They do not get the per-model thinking defaults of the reviewed list. If you set a thinking level in CodeGem, SuperQode assumes low, medium or high until that model is reviewed.

Neon Is Slow When Thinking Is High

Neon's default thinking level is high. That is Google's default, and CodeGem sends high unless you choose another level. On a hard problem that is the right starting point. For everyday coding it can feel slow, because the model reasons for a long time before the first visible text, even when streaming is on. Thinking is also billed as output.

Use low or medium for ordinary edits and questions. Keep high or xhigh for difficult, multi-step work. xhigh uses more tokens and takes longer, so use it when high is not enough. CodeGem rejects minimal for Neon.

This control lives on CodeGem, not on the simple Google API key connection. That simpler path does not send a thinking level, so Google's own default applies (high for Neon). The Mac Google connection screen has no thinking-level menu. To choose a level, use CodeGem and set model_policy.reasoning in a harness file:

YAML
model_policy:
  reasoning: medium

You can start from a copy of the built-in template:

Shell
superqode harness init my-codegem --template codegem --output codegem.yaml

Edit the file, then run it with --provider google and --model gemini-4-neon, or connect that harness in SuperQode.

Connect from the Mac App

Open a project in SuperQode. Choose API key, then Google under US Labs. SuperQode loads the reviewed Gemini list, plus any new models from the catalog, into the examples menu. You can also type a model id, such as gemini-4-neon, into the Model ID field.

Create a key in Google AI Studio. Paste it, leave Remember in Keychain on if you want the key saved in your Mac's Keychain, and click Use provider. Use Test model if you want a short, no-tool check. Google may charge for that request.

To use CodeGem instead of the default SuperQode harness, choose Connect a model to a harness, pick CodeGem, then connect Google with your API key the same way.

Connect from the Terminal

Set a key in your environment, or store one with superqode auth login google (saved in ~/.superqode/auth.json with 0600 permissions). Environment variables take precedence over the stored file.

Shell
export GEMINI_API_KEY=your-key
# or: export GOOGLE_API_KEY=your-key

superqode connect setup google
superqode models --provider google

Start SuperQode in your project and connect a model:

Terminal
:connect byok google gemini-4-neon

You can also run superqode connect byok google gemini-4-neon from the shell, or type :connect gemini-4-neon in the TUI and let SuperQode resolve the Google provider. :connect byok opens the provider and model picker. After you are connected, :model switch google/gemini-4-neon or :models set gemini-4-neon changes the model.

For CodeGem, start SuperQode, open :connect, choose Connect a harness with your model, then CodeGem, then BYOK for Gemini. :connect harness-codegem opens that flow directly. A headless run looks like this:

Shell
superqode harness run codegem "Explain this project and suggest one small fix"   --provider google --model gemini-4-neon --working-dir .

Pass a custom yaml file instead of codegem when you have set a thinking level there.

Get SuperQode

In the terminal:

Shell
uv tool upgrade superqode

If you do not have it yet, use uv tool install superqode. For the Mac app, use Download for Mac on the SuperQode site.

The CodeGem model guide has starting configurations for each reviewed model. The Google provider page and the BYOK guide cover the API key route.

πŸ“š Our blogs are also published on

Follow along wherever you already read

πŸ’‘ Found this helpful? Share it with your network and help others discover these insights!

Try it

Try Gemini 4 Neon in SuperQode

Download SuperQode for Mac, connect Google, and pick gemini-4-neon.