LLMsGaurav3 minNew

When to Use Gemini 4 Neon

When should you use Gemini 4 Neon instead of Gemini 4 Carbon or Hydrogen?

What Google shipped on November 18

Google introduced Gemini 4 Neon for coding, knowledge work, and science. On the post, the API price is $3 per million input tokens, $0.15 per million cached input tokens, and $15 per million output tokens, with a 1 million token window. Google says it is in the Gemini API, in Gemini for business, and for Pro and Ultra subscribers in Google Antigravity and the Gemini app. The November 19 briefing carries the release with its source.

Google places Neon between two siblings. Carbon is for the hardest, longest projects. Hydrogen is for fast everyday work. Neon is the one for complex jobs that still have to finish often.

A rank is not a reason to move the default

Google says Neon scores 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level, third overall, and 70.9 percent on the Vals Index, second. Those figures are Google's report of outside indexes, not a CuratedAI measurement. The same post says the default high thinking level costs 32 percent less per task than Carbon on that index. A lower cost per task can come from price, from fewer tokens, or from a faster stop. It does not say Neon wins your repository.

Match the model to how often the job runs

A hard job you repeat
Start with Neon. Coding, a long document, a spreadsheet, or a research brief that has to come back as something you can check. The 1 million token window is the reason to try it when the inputs do not fit a smaller call.
One long project
Keep Carbon. Google still describes Carbon as the model for the hardest, longest work. Neon's speed is not a substitute for a job that has to stay on one thread for hours.
Everyday turns
Stay on Hydrogen. Drafts, short edits, and routine agent steps do not need Neon's price. Google describes Hydrogen as the fast everyday model.
You already pay for Carbon
Replay a week of real tasks on Neon at the default high level, then again at xhigh only for the ones high gets wrong. Keep Carbon wherever Neon drops a condition or fails to finish.

Price the finished task

  • Save the thinking level with the run. High and xhigh are different products on the same name. Google's 61 is the xhigh score. The 32 percent claim is about the default high level.
  • Count omissions. A fluent answer that drops a cell, a citation, or a failing test is a miss. Cached input only helps when the same prefix is actually cached.
  • Keep the harness still. Neon inside Antigravity is not the same test as Neon on the API. Change one of those at a time. The model migration test is the longer version of this check.

Reviewed November 19, 2026, from Google's introduction post. For the neighboring choice on the OpenAI side, see when to use GPT-6.2 Astra or Sol.

Questions

When should you use Gemini 4 Neon instead of Gemini 4 Carbon or Hydrogen?
Use Neon for complex coding and knowledge work you run often. Keep Carbon for the longest projects, and Hydrogen for fast everyday tasks. Check the bill on your own sample before you switch.

Share

Use Neon for complex coding and knowledge work you run often. Keep Carbon for the longest projects, and Hydrogen for fast everyday tasks. Check the bill on your own sample before you switch.

Share on X LinkedIn Reddit Email
Copied