Key takeaways
- Google priced Gemini 4 Neon at $3 per million input tokens, $0.15 for cached input and $15 per million output tokens, with a 1 million token context window.
- Google said Neon finishes tasks for about 32% less than Gemini 4 Carbon at the default high thinking level and scored 61 on the Artificial Analysis Intelligence Index at xhigh.
- The model is rolling out to paid Gemini API developers, Google AI Pro and Ultra subscribers in the Gemini app and Antigravity, and business customers.
In this story 5 sections
Google introduced Gemini 4 Neon on Wednesday, Nov. 18, adding a third model that sits between the larger Gemini 4 Carbon and the quicker, cheaper Gemini 4 Hydrogen. Alphabet (NASDAQ: GOOGL) set the Gemini 4 Neon price at $3 per million input tokens and $15 per million output tokens, and said the model is meant for multistep work in coding, office tasks and science, according to Google's Gemini 4 Neon announcement.
Where Neon sits in the Gemini 4 family
Google now describes three tiers. Carbon is the model it recommends for the hardest and longest projects. Hydrogen, released about a week earlier, is built for fast everyday jobs, and our Gemini 4 Hydrogen pricing story covers that launch. Neon is the compromise between them: more depth than Hydrogen, at a lower cost per task than Carbon.
Google said Neon completes tasks for about 32% less than Carbon when both run at the default "high" thinking setting. Neon has four thinking levels, and the top one is called "xhigh." In Google Antigravity, the company's agent coding tool, Neon can also act as a coordinator that splits a job among several Hydrogen subagents working in parallel. That extends the multi agent features Google opened to developers earlier this month, described in our Antigravity teamwork article.
The price, and what it means per job
- Input: $3 per million tokens
- Cached input: $0.15 per million tokens, which Google called a 95% discount
- Output: $15 per million tokens
- Context window: 1 million tokens
Those rates are exactly double Hydrogen's $1.50 input and $7.50 output prices. In our own example, a request that reads 200,000 tokens and writes 20,000 would cost about $0.90 on Neon ($0.60 for input plus $0.30 for output), compared with about $0.45 on Hydrogen. If those 200,000 input tokens came from the cache, the Neon input charge would fall to $0.03.
Benchmarks Google cited
Google published a long list of scores, many from outside testing groups. These are the company's figures as stated in its post:
- Artificial Analysis Intelligence Index: 61 at the xhigh setting, which Google said ranks third overall
- Artificial Analysis Coding Agent Index: 58.6, first overall
- Vals Index: 70.9%, second place
- Terminal-Bench 4.1: 65.2%
- Zapier AutomationBench: 56.3%, first overall
- Andon Labs Vending-Bench 2: $16,843, first overall
- HLE-Diamond: 71.3%
- SEC-bench Pro: 90.5%
Google also said Neon ranked first on Legal Research Bench and Tax Agent Bench but did not give scores for those two. On cost, the company said Neon spends roughly 50% to 80% less per task than leading rival frontier models on the Vals and Agent Arena tests at the high setting.
Safety numbers
Google said Neon runs on the same Gemini 4 safety stack as Carbon, with comparable real time safeguards against cyber and biosecurity misuse but fewer needless refusals. Among the figures it shared were a 0.09% rate of misaligned actions on its agentic safety suite, a 99.2% pass rate on an agent honesty test and a 2.45% shortcut rate on a coding cheating evaluation, which Google said is the lowest of any Gemini 4 model. A security focused version, Neon for cyber, is going to vetted defenders in Google's Fairwind Program over the coming days, with access requested through a Google Cloud account team.
Who can use Gemini 4 Neon now
Google said the rollout started with the announcement:
- Developers on the paid tier of the Gemini API, through Google AI Studio and Android Studio
- Google AI Pro and Ultra subscribers, from the model menu in the Gemini app and in Google Antigravity
- Business customers, through the API, the model picker in Gemini for business and Antigravity for enterprises
The post does not mention Vertex AI. For details on the top tier, see our report on Gemini 4 Carbon pricing and availability, and follow more model releases on our AI page.
Share this story
Spotted an error? Tell us


