Gemini 4 Neon
gemini/gemini-4-neon
Back to model catalog

Gemini · Model guide

Gemini 4 Neon

Compare
gemini/gemini-4-neonReleased Nov 18, 2026

Gemini 4 Neon is Google’s Gemini 4 model for complex software engineering, knowledge work, and science workflows, adapted from Gemini 4 Carbon and positioned between Hydrogen and Carbon. It aims for frontier-level quality with fast end-to-end task completion and offers four thinking levels from low to xhigh, defaulting to high. It accepts text, images, audio, and video and returns text within a shared 1M-token input and output window.

Modalities
TextImageAudioVideoText
Context
1M
Max output
66K
Price / 1M tokens
$ 3.00in$ 15.00out
StreamingTool callingJSON outputImage inputThinking mode

What this model is good at

What the vendor positions it for, and which of our models to reach for instead.

VendorGoogle positions Neon between Hydrogen and Carbon for high-frequency complex tasks: frontier-level reasoning on complex coding, analytical, and enterprise workflows with leaner token use and faster end-to-end completion.source

Best for
  • Repository-level agentic coding · Google says it completes repository-level debugging and multi-file refactoring in fewer steps, and across long tool-heavy sessions reduces redundant file reads and handles context compaction reliably.(vendor claim)source

  • Analytical knowledge work · Google highlights synthesizing dense documents, spreadsheets, charts, and multi-step tool outputs into structured deliverables, with strong legal and finance results.(vendor claim)source

  • Science and quantitative work · Google lists science workflows and quantitative, analytical reasoning among the tasks it is built for.(vendor claim)source

  • Secure coding and vulnerability patching · Google says it writes code with fewer security vulnerabilities and improves root-cause vulnerability patching, and lists vulnerability finding, validation, and patching among its intended uses.(vendor claim)source

Neighbouring models
  • gemini/gemini-4-carbonstronger —Choose Gemini 4 Carbon for the hardest architecture, debugging, science, and security problems.(vendor claim)source
  • gemini/gemini-4-hydrogenfaster —Keep Gemini 4 Hydrogen as the default for everyday coding and agentic workflows, where Google positions it for fast everyday execution.(vendor claim)source
  • gemini/gemini-3.5-flash-litefaster —Choose Gemini 3.5 Flash-Lite for high-volume, low-latency tasks.(vendor claim)source
Compare them side by side →

Pricing and billing

Billed per token. Cached input is charged at the cache-read rate.

Price / 1M tokensYou pay
Input$ 3.00
Output$ 15.00
Cache read$ 0.15
Cache write$ 3.00
Estimate a request
$ 0.0600
Estimated cost per request
10.0K × $ 3.00
+ 2.0K × $ 15.00
≈ $ 60.00 per 1,000 requests

Capabilities and limits

What CrossModel guarantees across every route this model can take right now.

Context window
1.0M tokens
Max output
65.5K tokens
Input / output modalities
Text + Image + Audio + Video → Text
Streaming
Supported
Tool calling
Supported
Structured output (JSON)
Supported
Image input
Supported
Thinking mode
Supported· always on
Reasoning effort
low · medium · high
Available endpoints
/v1/chat/completions · /v1/responses · /v1/messages

Published benchmarks

Scores the vendor reported, with the evaluation setup each one came from.

Agentic work
  • Agents' Last Exam v1.1
    score·Google DeepMind · 2026-11-18effort xhighAntigravity-CLI (agy) 1.4.8setup not fully disclosed

    Google self-computed averaged rubric score on the 122-task v1.1 set with the Antigravity-CLI (agy) 1.4.8 harness, five-hour window, safety filters on. Argon and Carbon were run with ALE-Claw instead. Same table: Claude Opus 5.5 71.6% (public leaderboard), GPT-6.2 Sol 73.0% (its system card). Methodology: https://deepmind.google/models/evals-methodology/gemini-4-neon

    75.8%
Coding
  • SWE-sweep
    score·Google DeepMind · 2026-11-18effort xhighAntigravity-CLI (agy) 1.5.9setup not fully disclosed

    Google self-computed with the public Antigravity-CLI (agy) 1.5.9 harness, whose default sub-agent fan-out tool suits bug-finding tasks; reports the subtask-weighted reward. Highest thinking setting, single attempt. The same card reports 12.8% with a mini-swe-agent vision harness, the setting comparable with Argon 4.3% and Carbon 9.6%; Hydrogen's 13.5% used agy 1.4.8. Same table (provider harnesses): Claude Opus 5.5 10.1% (Claude Code), GPT-6.2 Sol 8.3% (Codex). Methodology: https://deepmind.google/models/evals-methodology/gemini-4-neon

    39%
Long context
  • GraphWalks BFS 256K to 1M
    f1·Google DeepMind · 2026-11-18effort xhighsetup not fully disclosed

    Google self-computed F1 on the 200 problems with context lengths from 256K to 1M tokens; highest thinking setting. Same table: Claude Opus 5.5 66.8% (matches the Carbon card), GPT-6.2 Sol 80.1%. Full reproduction settings not disclosed. Methodology: https://deepmind.google/models/evals-methodology/gemini-4-neon

    95.2%
Multimodal reasoning
  • Chartography (no tools)
    score·Google DeepMind · 2026-11-18effort xhighsetup not fully disclosed

    Google self-computed without tools following the official Surge public leaderboard; highest thinking setting. Same table: Claude Opus 5.5 66.3% (its system card), GPT-6.2 Sol 75.1% (Google self-computed). Methodology: https://deepmind.google/models/evals-methodology/gemini-4-neon

    73.5%
Scientific research
  • Terminal-Bench-Science 0.1
    accuracy·Google DeepMind · 2026-11-18effort xhighmini-swe-agent (vision)5 runssetup not fully disclosed

    Google self-computed with a mini-swe-agent vision harness, average reward over 5 attempts per task, highest thinking setting. Same table: Claude Opus 5.5 63.3%, GPT-6.2 Sol 68.0% (public leaderboard); the Opus figure differs from the 58.7% in its own system card. Methodology: https://deepmind.google/models/evals-methodology/gemini-4-neon

    70.5%
Video understanding
  • LVBench (static)
    accuracy·Google DeepMind · 2026-11-18effort xhighsetup not fully disclosed

    Google self-computed without tools (vision-only); Gemini video sampled at 1 FPS, low resolution. GPT-6.2 Sol used 500 frames and Claude Opus 5.5 600 frames due to API limits. Same table: Claude Opus 5.5 83.7%, GPT-6.2 Sol 87.6%. Full reproduction settings not disclosed. Methodology: https://deepmind.google/models/evals-methodology/gemini-4-neon

    89%

Vendor-reported numbers, not CrossModel measurements. Scores are only comparable when the benchmark version, metric and evaluation setup match, so nothing here is averaged or ranked.

Use it in your tools

Point the base URL at CrossModel and paste this model ID — every tool below has a setup guide.

See all integration guides →

Frequently asked questions

What is Gemini 4 Neon?
Gemini 4 Neon is available on CrossModel as gemini/gemini-4-neon. It has a 1M-token context window and can return up to 66K tokens per request.
How much does Gemini 4 Neon cost?
Currently $ 3.00 per 1M input tokens and $ 15.00 per 1M output tokens. Cached input is billed at $ 0.15 per 1M.
Does Gemini 4 Neon support tool calling and structured output?
Tool calling: Supported · Structured output (JSON): Supported · Image input: Supported · Streaming: Supported · Thinking mode: Supported
Which endpoint do I call?
For text models, OpenAI Chat Completions (/v1/chat/completions), OpenAI Responses (/v1/responses) and Anthropic Messages (/v1/messages) all work — pick whichever your SDK already speaks.
Can I try it without writing code?
Yes. Sign in and open the Playground on this page to chat with the model directly — no API key needed. Usage is billed from your wallet at the normal rate.

Chat first, integrate later

Chat first, then wire it in once you like the answers.