---
title: "Google DeepMind launches Gemini 4 Neon, a faster mid-tier model for coding, analysis and agents"
url: https://llmreport.gr/news/google-deepmind-launches-gemini-4-neon-a-faster-mid-tier-model-for-coding/
language: en
published: 2026-11-18T23:11:55.246Z
updated: 2026-11-18T23:11:55.251Z
site: "LLM Report"
section: "Models"
---

# Google DeepMind launches Gemini 4 Neon, a faster mid-tier model for coding, analysis and agents

Google DeepMind released Gemini 4 Neon on 18 November 2026 as a quicker, cheaper option between Carbon and Hydrogen. It reaches developers, businesses and paying Gemini app users, and a cyber variant is for vetted defenders.

> Written with AI from the sources listed below. Editorial responsibility: Konstantinos Perros. How we write: https://llmreport.gr/how-we-write/

## Key facts

- Google DeepMind announced Gemini 4 Neon on 18 November 2026 and began rolling it out the same day. [1]
- Google lists the price at $3 per million input tokens and $15 per million output tokens, with cached input at $0.15 per million. [1]
- Google says Neon scored 61 on the Artificial Analysis Intelligence Index at xhigh thinking, placing third overall. [1]
- Neon is offered in Google AI Studio, Android Studio, Gemini for business, Google Antigravity and the Gemini app, with some surfaces limited to paid Pro and Ultra plans. [1]
- A cyber-focused version, Gemini 4 Neon for cyber, goes to trusted defenders in the Fairwind Program over the following days. [1]

## What Google released and how to get it

Google DeepMind introduced Gemini 4 Neon on 18 November 2026. The company describes it as a model that brings frontier-level intelligence to complex software engineering, knowledge work and science workflows while finishing tasks quickly from start to end.

On pricing, Google states that input tokens cost $3 per million, while output tokens cost $15 per million. Cached input is cheaper at $0.15 per million, which the company calls a 95% discount. The context window holds 1 million tokens.

Developers on the paid tier of the Gemini API can use Neon through Google AI Studio and Android Studio. Businesses can select it inside Gemini for business, either by API or from the model picker, and in Google Antigravity. Paid Google AI Pro and Ultra subscribers can try it in Antigravity, and Pro and Ultra subscribers can also pick it in the Gemini app. Google says rollout starts on the day of the announcement.

## Benchmark claims and cost positioning

All figures below are Google's own reporting of results, some of which it attributes to third-party evaluators. At the xhigh thinking level, Google says Neon is ranked third overall on the Artificial Analysis Intelligence Index with a 61, just behind larger and slower models. It also places second on the Vals Index at 70.9%, and second on the Agent Arena Index with 1336 points across more than 4,200 real-world agent sessions. Google adds that Neon is ranked second for steerability, meaning how well it adjusts to corrections made mid-task.

For enterprise tasks, Google says Neon is first on the Legal Research Bench and the Tax Agent Bench, scores 73.5% on Chartography and 75.8% on Agents' Last Exam v1.1, reaches 56.3% on Zapier's AutomationBench and $16,843 on Andon Labs' Vending-Bench 2. In science and maths, it reports 70.5% on Terminal-Bench Science, 84.0% on RiemannBench and 71.3% on HLE-Diamond, calling each best in class.

For coding, Google reports first place on the Artificial Analysis Coding Agent Index at 58.6, 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at the default high setting. On SWE-sweep it says Neon leads the mini-SWE harness at 12.8% and reaches 39.0% inside Google Antigravity. The company positions the model for multi-file debugging and refactoring with narrow, targeted edits, and as an orchestrator that directs parallel Hydrogen subagents in Antigravity.

On cost, Neon offers four thinking levels so teams can match spend to the workload. At the default high level, Google says cost per task is 32% below Carbon's while keeping a high Intelligence Index score. It also puts Neon on the Pareto frontier of the Vals Index and Agent Arena, at roughly half to one-fifth the per-task cost of top rival models. At xhigh, it says costs remain at least 36% below competitors at their highest settings.

## Cyber variant and safety measures

Gemini 4 Neon for cyber is a version aimed at defensive security. Google says it writes more secure code than Carbon at every difficulty on AutoBaxBench, with 76% on the hard tier. It reports 51.26% on WebExploitBench v1.2 at half the price of competing large models, and 90.5% on SEC-bench Pro for finding vulnerabilities at about half the cost per challenge. For exploit validation on ExploitGym, Google says it matches frontier-level results with half of Carbon's token use, and on Collinear's CWE-bench v2 for patching it reaches 70.7%, a 3.4% improvement over Carbon.

Access is restricted. Eligible members of the Fairwind Program are due to receive the cyber model over the next few days, and organisations can request entry via their Cloud account team.

On safety, Google says Neon keeps the protections against severe misuse and prompt injection that Carbon has, with real-time safeguards for cyber and biosecurity risks. The company reports a refusal rate of 1.98% on Terminal-Bench 4.0 and zero on Terminal-Bench-Science. It also says the model took shortcuts in 2.45% of cases on its coding cheating evaluation, had a 0.09% misaligned action rate on its agentic safety suite, and passed 99.2% of its agent honesty evaluation. On Gray Swan's indirect prompt injection benchmark, Google says Neon with safeguards at the highest thinking level resisted every attack. A model card has the details.

## Where Neon fits in the Gemini 4 line and the app

Google places Neon between two siblings: Carbon for the hardest long-horizon projects and Hydrogen for quick everyday execution. Neon is meant to balance depth and efficiency for complex, frequent workflows.

In the Gemini app, Google says Neon writes with clearer structure, reads spreadsheets, PDFs and photos more precisely, and can draw on Gmail and Docs. It also pulls in web sources for fast-moving topics. Tasks in Gemini can now run on Neon, which Google says completes multi-step jobs faster than larger models while stretching usage limits, with examples such as analysing billing statements or cross-checking receipts for a refund claim.

## Sources

1. Google DeepMind: [Introducing Gemini 4 Neon](https://deepmind.google/blog/introducing-gemini-4-neon)
