AI NewsModels & agentsAnnouncement
Google releases Gemini 4 Neon at $3 per million input tokens
Google launched Gemini 4 Neon at $3 per million input tokens. Output is $15 per million, with a 1 million token context window.

Why it mattersA coding agent model at $3 per million input tokens is a direct option against Claude Opus and GPT-6.2 Sol. The cached rate drops to $0.15.
A new coding model from Google arrived on 18 November 2026. The price sits well below Claude Opus and GPT-6.2 Sol. Google is pointing the model at long-running coding agents.
Google released Gemini 4 Neon with a 1 million token context window. Input tokens cost $3 per million. Output tokens cost $15 per million. Cached input tokens cost $0.15 per million, which Google calls a 95 percent discount. The company says the model is built for "fast end-to-end task completion across complex software engineering."
Where Google says the model lands on public benchmarks
Google lists six published benchmark results in the post. On the Artificial Analysis Intelligence Index, it scores 61 at the xhigh thinking level. That ranks third overall. On the Vals Index it ranks second with 70.9 percent. On the Artificial Analysis Coding Agent Index, Google says the model is first overall at 58.6. On Terminal-Bench 4.1 it scores 65.2 percent. On FrontierCode 2.0 it scores 46.2 percent at the default high setting. On the Agent Arena Index it ranks second globally with 1336.
Every one of those numbers is the vendor reporting a figure from a third-party benchmark on the day of release. Read them the way you would read any launch-day score.
What Google says about hallucination
Google writes that the model recorded its "lowest fabrication rate yet" in internal testing. The testing ran inside Google Antigravity, the company's agent IDE. The post gives no percentage for that figure. A buyer who cares about hallucination should read the sentence as a signal that Google measured something. The next step is to ask for the number in a sales call.
Where you can use it today
Gemini 4 Neon is rolling out in the Gemini API through Google AI Studio and Android Studio. It is in Gemini for Business through the API and the model dropdown. Paid Google AI Pro and Ultra subscribers get it inside Google Antigravity. Pro and Ultra subscribers also see it inside the Gemini app. Cybersecurity teams in the Fairwind Program get access through that program.
The model has four thinking levels. Google names the default as "high" and the top setting as "xhigh". A developer tunes the setting per task. A routine edit does not have to pay for the deep reasoning that a hard bug needs.
For a team running an agent on every change, input tokens dominate the bill. The repository, the diff, the recent commits and the open issue all enter the context on every turn. A model that reads a million tokens per call at $3 per million changes the arithmetic. Reusing most of that context from the cache costs 5 percent of the list price. An agent that was too expensive to run overnight last month may be back on the table this week.
Source
- Google, Introducing Gemini 4 Neon
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
