Google releases Gemini 4 Neon: a mid-priced model for coding and knowledge work
Gemini 4 Neon sits between Hydrogen and Carbon at $3/$15 per million tokens. Where it's available, what it can do, the caveats, and how to test it against your current model.
- Author
Umair Mateen- Published
- Reading time
- 5 min
- Category
- AI
Ideas in this post
Tap one for a 5-second answer. Underlined words in the article work the same way.
0 of 4 understood

Google released Gemini 4 Neon on 18 November 2026, a new model that sits between its fast Gemini 4 Hydrogen and its deep-reasoning Gemini 4 Carbon. It is aimed at software engineering, knowledge work and science tasks, and it is rolling out to developers, businesses and paying Gemini subscribers at $3 per million .
What it is
Google pitches Neon as the middle option in the Gemini 4 line: more capable than Hydrogen, which is built for quick, repeated steps, and cheaper and faster than Carbon, which is meant for long, deep work. In Google's words, "Neon is trained to complete multi-step work without cutting corners."
The main specs, from Google's announcement and its Google Cloud model page:
Quick check · Myth or fact
Tokens: myth or fact?
Decide for yourself first, then tap each statement to check it.
0 of 3 checked
From the explainer What are tokens in AI? Explained with word-game tiles
- : about 1 million tokens, shared between input and output. A token is roughly three-quarters of a word.
- Inputs: text, images, audio, video, PDFs and Office or OpenDocument files. Output is text.
- : four settings, Low, Medium, High (the default) and Extra high, that trade speed and cost for more careful reasoning.
- Tools: , structured output, code execution, Google Search grounding, URL reading and computer use are listed as supported. Fine-tuning is not.
What it costs
Google lists API prices of $3 per million input tokens, $15 per million output tokens and $0.15 per million cached input tokens, a 95% discount when the same long prompt or document is sent repeatedly. Remember that "thinking" counts as output, so the higher thinking levels raise the bill.
Google also says that at its default High setting, Neon completes tasks at a 32% lower cost per task than Carbon. That is Google's own measurement; your mix of prompts may give a different result.
Where you can use it
- Developers: the paid tier of the Gemini API in Google AI Studio and Android Studio.
- Businesses: Gemini for business, through the API and the model picker, and Google Cloud's Gemini Enterprise Agent Platform under the model ID
gemini-4-neon. - Subscribers: Google AI Pro and Ultra plans, in the Gemini app and in Google Antigravity, Google's coding agent tool. Google's Antigravity note says Free and AI Plus plans are not included there.
In Antigravity, Google's launch note says Neon becomes the default model for new conversations on plans that include it. It also adds an "Orchestrator (Neon)" mode, where Neon plans and reviews the work while the cheaper Hydrogen makes the code edits and runs tests, handing back to Neon if it gets stuck. That mode is for Pro and Ultra subscribers, not enterprise accounts.
How good is it
Google reports strong scores on coding and agent benchmarks, including first place on Artificial Analysis's Coding Agent Index and 65.2% on Terminal-Bench 4.1, plus low rates of shortcut-taking and misaligned actions in its own safety tests. These are vendor-reported figures; independent results usually take a few weeks to settle, and benchmark rankings rarely predict how a model handles your own documents and code.
Limits and caveats
- Context differs by product. The API offers about 1 million tokens, but Google's Antigravity note says up to 512,000 tokens there.
- Some settings are rejected. Google's developer guide and model page say custom temperature and similar sampling settings are ignored, and the older thinking-budget setting returns an error. Code written for earlier Gemini models may need small changes.
- No explicit caching. Caching happens automatically for repeated prompt prefixes of 4,096 tokens or more; you can't manage caches yourself.
- Rollout, not a switch. Google describes the launch as a rollout, so access may arrive at different times across accounts and regions. A security-focused version for vetted defenders is coming separately over the next few days.
Who should care
Teams that run coding agents, document-heavy workflows or customer-facing assistants on Gemini now have a middle price point. If Carbon felt too slow or costly and Hydrogen too shallow for your harder tasks, Neon is the model to test. Teams on other providers can treat it as one more option to compare on price per completed task, not price per token.
What to do this week
- Pick 20 to 50 real tasks from your own workload and run them on Neon at High, and on the model you use today. Compare quality and total cost per finished task.
- Check your code for temperature or thinking-budget settings that Neon ignores or rejects.
- If you repeat long system prompts or documents, structure them as a shared prefix so the automatic cache discount applies.
- If your team uses Antigravity, note that new conversations may now default to Neon, which can change both output and usage.
The DevDrift helps teams test and switch AI models inside their workflows with approvals and audit trails in place.
Sources
You’ve read the news. Now understand it.
- Learn · 5 min · Myth or factWhat are tokens in AI? Explained with word-game tilesTokens are the small pieces of text an AI model reads and writes, and the unit used to measure limits and cost. Explained with a word game whose tiles carry chunks of words.Start in 5 seconds →
- Learn · 5 min · Try the dialWhat is a context window? Explained with a clerk's deskA context window is how much text an AI model can look at in one go: your instructions, documents, the chat so far and its own reply. Explained with a clerk working at a desk of fixed size.Start in 5 seconds →
- Learn · 5 min · Try the dialWhat are reasoning models? Explained with a clerk's scrap paperA reasoning model is an AI model that works through a problem in hidden steps before answering. Think of a clerk who works a sum out on scrap paper, then writes one clean figure on the form.Start in 5 seconds →
- Learn · 5 min · Put it in orderWhat is tool use in AI? Explained with a calculator in an examTool use, or function calling, lets an AI model ask your software to run a tool, like a lookup or a calculator, and use the result. Explained with a student allowed a calculator in an exam.Start in 5 seconds →


