For Developers/Models/Gemini 4 Neon
Closed SourceGoogleReleased 2026-11

Gemini 4 Neon

Google's middle Gemini 4 model, between Carbon and Hydrogen, for complex coding and enterprise work

Context window

1M (shared input + output)

Input / 1M tokens

$3.00

Output / 1M tokens

$15.00

Provider

Google

Paid tier only: $3 input / $15 output per 1M tokens (output includes thinking tokens), flat across the full 1M context, with no introductory price. Cached input $0.15 (implicit caching only; explicit caching is not supported). Batch and Flex: $1.50 / $7.50 (cached $0.075). Priority: $5.40 / $27. Grounding with Google Search or Maps: 5,000 free requests a month shared across Gemini 3 and newer models, then $14 per 1,000. Not on the free tier. · Data verified 2026-11-20

Gemini 4 Neon (November 18, 2026) is a Gemini 4 model adapted from Gemini 4 Carbon that Google places between Carbon, for the hardest long-horizon projects, and Hydrogen, for fast everyday execution. Google built it for complex coding, analytical reasoning and production enterprise workflows. It costs $3/$15 per 1M tokens (cached input $0.15) across a 1,048,576-token window shared by input and output, writes up to 65,536 tokens per synchronous response or up to 1M with long decode, and has a March 2026 knowledge cutoff. Google reports 61 on the Artificial Analysis Intelligence Index at xhigh (third overall, ahead of Claude Opus 5.5 at 59 and GPT-6.2 Sol at 57), first place on the Artificial Analysis Coding Agent Index (58.6), and a 32% lower cost per task than Carbon at the default high thinking level. It is on the paid Gemini API tier and available to Google AI Pro and Ultra subscribers in the Gemini app and Google Antigravity. According to 9to5Google, the Gemini app picker labels it "Deeper analysis" and it took Gemini 4 Argon's place there; Google's own posts do not mention Argon.

How to access it

Generally available since November 18, 2026 on the paid Gemini API tier (model ID gemini-4-neon) in Google AI Studio and Android Studio, with Batch, Flex and Priority processing; not on the free tier. Businesses get it in Gemini for business (API or model drop-down) and Google Antigravity. Google AI Pro and Ultra subscribers can pick it in the Gemini app model drop-down and use it in Google Antigravity (desktop app, CLI and IDE extensions, including Orchestrator). A cyber variant is rolling out to eligible Fairwind Program members.

Strengths

  • ✓61 on the Artificial Analysis Intelligence Index v5.0.1 at xhigh, third overall; Claude Opus 5.5 scores 59 and GPT-6.2 Sol 57 (Google model card)
  • ✓First on the Artificial Analysis Coding Agent Index (58.6) and on Zapier's AutomationBench (56.3%), per Google
  • ✓32% lower cost per task than Gemini 4 Carbon at the default high thinking level (Google-reported)
  • ✓95.2% on GraphWalks BFS at 256K to 1M tokens vs 66.8% for Claude Opus 5.5 and 80.1% for GPT-6.2 Sol (Google model card)
  • ✓90.5% on SEC-bench Pro and 70.7% on CWE-Bench v2 (Google-reported)
  • ✓Flat $3/$15 pricing across the 1M context, with no introductory rate that expires

Best for developers who...

Repo-level edits, debugging and refactoring that need more depth than Gemini 4 HydrogenEnterprise knowledge work in finance, legal and spreadsheets (Google reports first place on Legal Research Bench and Tax Agent Bench)Planning in Google Antigravity's Orchestrator, with Gemini 4 Hydrogen running routine stepsAnalysis over very long documents and codebases

When to choose it (and when not to)

Reach for Gemini 4 Neon when...

  • →Coding and analysis that need more depth than Gemini 4 Hydrogen, at lower cost per task than Carbon
  • →Frontier-level work from December 1, 2026, when Carbon and Argon rise to $4/$20 and Neon stays at $3/$15
  • →Long-context work: Google's model card shows its widest lead on GraphWalks at 256K to 1M tokens
  • →Antigravity setups where Neon plans and Hydrogen executes

Look elsewhere if...

  • ✕High-volume everyday tasks: Gemini 4 Hydrogen costs half as much and has a free tier
  • ✕The hardest long-horizon projects, where Google points to Gemini 4 Carbon
  • ✕Price-per-token decisions before December 1, 2026: Carbon and Argon cost $2/$10 until November 30
  • ✕Terminal and repository coding where Claude Opus 5.5 is level or slightly ahead in Google's own table (Terminal-Bench 4.1, FrontierCode 2.0)
  • ✕You need explicit context caching (implicit only) or the free API tier

How to use it

  • ›Thinking defaults to high; Google reports its top Intelligence Index score at xhigh, and low or medium cut cost
  • ›Caching is implicit only, so keep a stable prompt prefix to get the $0.15 cached rate
  • ›Raise hardcoded output caps to 65,536 tokens and use long decode for longer outputs

Quickstart

Python
from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-4-neon",
    input="Refactor this module and explain each change: ...",
)
print(interaction.output_text)

Install with pip install -U google-genai. Pattern from Google's Gemini API quickstart (Interactions API); needs a paid-tier API key.

API model id: gemini-4-neon

Benchmarks

BenchmarkScoreNotes
Artificial Analysis Intelligence Index v5.0.161 (xhigh)Google model card, third overall; Claude Opus 5.5 59, GPT-6.2 Sol 57. Artificial Analysis lists Gemini 4 Argon at 50
Artificial Analysis Coding Agent Index58.6Ranked first, per Google
Terminal-Bench 4.165.2%Google model card, cheating-adjusted accuracy; Claude Opus 5.5 65.8%, GPT-6.2 Sol 61.5%
FrontierCode 2.046.2% (high)Google model card, main score; Claude Opus 5.5 47.3%, GPT-6.2 Sol 43.8%
CursorBench 4.165.4%Google model card; Claude Opus 5.5 58.1%
SWE-sweep39.0%Google model card, Antigravity harness; 12.8% with the mini-SWE harness (Claude Opus 5.5 3.6%, GPT-6.2 Sol 11.6%)
Agents' Last Exam v1.175.8%Google model card; Claude Opus 5.5 71.6%, GPT-6.2 Sol 73.0%
AutomationBench56.3%Google model card, ranked first on Zapier's leaderboard; Claude Opus 5.5 42.5%, GPT-6.2 Sol 41.3%
Vals Index v2.270.9%Google model card, ranked second; Claude Opus 5.5 70.1%, GPT-6.2 Sol 66.1%
Vending-Bench 2$16,843Final bank balance, ranked first on the Andon Labs leaderboard, per Google
Terminal-Bench-Science 0.170.5%Google model card; Claude Opus 5.5 63.3%, GPT-6.2 Sol 68.0%
HLE-Diamond71.3%Google model card; Claude Opus 5.5 67.9%, GPT-6.2 Sol 64.0%
GraphWalks BFS 256K-1M95.2%Google model card, F1; Claude Opus 5.5 66.8%, GPT-6.2 Sol 80.1%
SEC-bench Pro90.5%Google model card; GPT-6.2 Sol 88.3%
Chartography73.5%Google model card; Claude Opus 5.5 66.3%, GPT-6.2 Sol 75.1%

Source: Gemini 4 Neon model card

Tools powered by Gemini 4 Neon

Compare Gemini 4 Neon

Compare Gemini 4 Neon with any other model

Build a comparison →
All model comparisons →

Learn the concepts