Gemini 4 Neon vs. Carbon: API Prices Flip in December

Gemini 4 Neon reaches paid users and developers. Its API costs more per token than Carbon through November, then less when Carbon’s introductory pricing ends.
Google's launch artwork for Gemini 4 Neon
Gemini 4 Neon launch artwork. Image: Google.

Google began rolling out Gemini 4 Neon on November 18, bringing a new reasoning model to paid Gemini app subscribers and developers building with its API. It is designed for coding, analysis and multi-step work, with a one-million-token context window and four selectable reasoning levels.

For developers comparing it with Google’s more powerful Gemini 4 Carbon, the price has an unusual wrinkle: Neon costs more per token until November 30, when Carbon’s introductory rates expire. From December 1, the published standard API rates make Neon cheaper per token. Actual task costs can still differ because the models do not necessarily use the same number of tokens.

Who can use Gemini 4 Neon?

Google is rolling Neon out to Google AI Pro and Ultra subscribers in the Gemini app, where eligible users can select it from the model menu. It is also coming to Gemini for business and to paid Pro and Ultra users of Google Antigravity, the company’s development environment. Google’s launch announcement lists access through the paid Gemini API and Google AI Studio as well.

The API and app are different products: a consumer subscription is not the same as per-token API billing. Google’s API price page lists no free API tier for Neon. Access and usage limits can vary by product and plan; a rollout announcement does not mean every account will see the model immediately.

Google positions Neon between Hydrogen, its faster everyday model, and Carbon, which it reserves for the most demanding long-running work. Neon can take text, images, audio and video as inputs and produces text. A million-token context is a capacity limit, not a guarantee that the model will correctly retrieve or reason over every detail in a long document.

Neon vs. Carbon: the price changes in December

The table compares standard Gemini API rates in U.S. dollars per million tokens. It excludes caching, tool charges and any differences in the number of tokens each model uses.

Model Input Output, including thinking
Neon $3 $15
Carbon through November 30 $2 $10
Carbon from December 1 $4 $20

For the same uncached token volumes, Neon’s rates are 50% higher than Carbon’s current promotional rates, but 25% lower than Carbon’s scheduled December rates. For example, one million input tokens and 100,000 output tokens would cost $4.50 with Neon, $3 with Carbon during the promotion, or $6 with Carbon afterward. These examples exclude other charges and assume identical token usage.

Neon’s output rate includes thinking tokens, so the visible answer alone does not tell the whole billing story. Google lists cached input at $0.15 per million tokens for Neon, with implicit caching only; explicit caching is not supported. Grounding and other tools can carry separate charges. Check the current price page and usage data before budgeting a production workload.

What independent tests show

Neon offers low, medium, high and xhigh thinking levels. More reasoning can improve performance on difficult work, but it can also increase latency and token use. In Artificial Analysis’s measurements, Neon scored 44 at low and 61 at xhigh on its Intelligence Index; average time per index task rose from 3.7 to 21.5 minutes. Those figures describe that test suite, not a promise for an individual prompt.

In a separate comparison with Carbon, the evaluator found Neon at xhigh used about 15% more output tokens per task, but cost roughly 17% less and completed tasks about 40% faster using Carbon’s non-discounted prices. That cost comparison uses Carbon’s future standard rate, not its lower November promotional rate. The benchmark also compares Neon at xhigh with Carbon at high, so it should not be treated as a universal ranking for every task.

How to decide which model to use

In the Gemini app, eligible subscribers can try Neon from the model selector and compare answers on their own documents or questions. Developers can select gemini-4-neon in a paid API project and test representative tasks at more than one thinking level. Record output quality, total billed tokens, latency and any tool charges; repeat the test after Carbon’s promotional price ends.

For coding agents, include tests that check whether proposed changes work and whether the model follows instructions across multiple files. Review generated code and sensitive actions before deployment. A benchmark score or a large context window does not replace those checks.

Leave a Reply

Your email address will not be published. Required fields are marked *

Previous Post
A laptop workspace representing cloud productivity apps, email, documents, and team collaboration

Google Docs Adds Figure Captions and Cross-References: How to Keep Them Updated

Related Posts