Key takeaways

  • Google released Gemini 4 Neon on November 18 in the Gemini API, Antigravity, the Gemini app for Pro and Ultra subscribers, and GitHub Copilot.
  • It costs $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window and no free API tier.
  • Google says Neon completes tasks for 32% less than Gemini 4 Carbon, whose API price doubles on December 1.

Google released Gemini 4 Neon on November 18, a mid-sized model in the Gemini 4 family built for coding agents, office work and science tasks. In the announcement, Google says Gemini 4 Neon is meant for "fast end-to-end task completion" and costs $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window.

Neon sits between its two siblings. Google pitches Carbon for the hardest long projects and Hydrogen for quick everyday work. Neon is the one Google wants developers to run all day: it claims Neon finishes a task for 32% less than Carbon at the default setting. That matters because Carbon's own price doubles on December 1.

Where you can use Gemini 4 Neon

The model is rolling out now in several places at once:

  • Gemini API, through Google AI Studio under the model name gemini-4-neon, and in Gemini for business.
  • Google Antigravity, Google's agent coding app, for paid Google AI Pro and Ultra subscribers.
  • The Gemini app, for Pro and Ultra subscribers, from the model menu. Google says it reads spreadsheets, PDFs and photos more precisely and runs Tasks faster than the larger models.
  • GitHub Copilot. GitHub made it generally available the same day on Copilot Pro+, Max, Business and Enterprise, in VS Code, Visual Studio, JetBrains IDEs, Xcode, the Copilot CLI and the coding agent. It is billed at Google's list price under usage-based billing.

Neon is not on the free tier of the Gemini API, according to Google's pricing page.

Gemini 4 Neon pricing against Carbon and Hydrogen

Google's prices per million tokens, as listed on November 18, 2026:

Model Input Output Cached input Batch input / output
Gemini 4 Neon $3.00 $15.00 $0.15 $1.50 / $7.50
Gemini 4 Carbon (until Nov 30) $2.00 $10.00 $0.10 $1.00 / $5.00
Gemini 4 Carbon (from Dec 1) $4.00 $20.00 $0.20 $2.00 / $10.00
Gemini 4 Hydrogen $1.50 $7.50 $0.075 $0.75 / $3.75

So for the next two weeks Neon costs more per token than Carbon. From December 1 it costs less. Cached input gets a 95% discount, which helps agents that resend the same files many times.

Google's cost claim is per task, not per token. It says Neon uses fewer steps and fewer redundant file reads, and that at its default "high" thinking level it costs 50% to 80% less per task than rival frontier models on the Vals and Agent Arena tests. Neon has four thinking levels.

What the benchmark numbers say

Google lists a long set of scores. A few that stand out:

  • Artificial Analysis Intelligence Index: 61 at the highest thinking level, third overall, according to Google.
  • Artificial Analysis Coding Agent Index: 58.6, first overall.
  • Terminal-Bench 4.1, which tests agents working in a terminal: 65.2%.
  • Vals Index: 70.9%, second, and first on Vals' legal research and tax agent tests.
  • Zapier AutomationBench: 56.3%, first.

Most of these are third-party leaderboards, but the numbers here are as Google reports them, on the settings Google chose. Google also says Neon made things up less often than any model it has tested in internal Antigravity runs, a claim with no outside check yet. A comparison of the benchmark results sets Neon beside its siblings and rivals.

On safety, Google says Neon refused 1.98% of tasks on Terminal-Bench 4.0, took shortcuts on 2.45% of tasks in its coding cheating test, the lowest of any Gemini 4 model, and passed 99.2% of an agent honesty test. Details are in the model card.

A cyber version for defenders

A version tuned for security work is coming "in the coming days" to members of Google's Fairwind Program for cyber defenders, through their Google Cloud account teams. Google reports 90.5% on SEC-bench Pro and 51.26% on WebExploitBench v1.2, at about half the cost of competing large models. It follows the Carbon release for cyber defenders in October.

What to watch

The Decoder counts Neon as Google's fourth new Gemini model in a month. The useful test comes over the next few weeks: independent runs on SWE-bench and Terminal-Bench, and whether developers in Copilot and Antigravity pick it over Carbon once Carbon's price doubles on December 1.

Sources

  1. Introducing Gemini 4 NeonGoogle
  2. Gemini Developer API pricingGoogle
  3. Gemini 4 Neon model cardGoogle DeepMind
  4. Gemini 4 Neon in GitHub CopilotGitHub
  5. Google ships Gemini 4 Neon, its fourth new Gemini model in a monthThe Decoder

Add ClaudeGPT Si as a preferred source on Google