ModelBrief · Gemini 4 Neon · Topic Page
Last updated Nov 20, 2026

Gemini 4 Neon Access and API Guide: Model ID, Thinking Tiers, Long Decode and Caching Rules

Published Nov 19, 2026 · Last updated Nov 19, 2026

Gemini 4 Neon shipped on all channels on launch day: the paid API, AI Studio, Gemini for business, Antigravity and the Gemini app — five routes at once, no waitlist. But every route has a bar: the API has no free tier, and the app and Antigravity need a Pro subscription at minimum. This guide covers all five routes, the model ID, thinking tiers, long decode and caching rules — and flags which claims are not yet officially confirmed.

API: model ID and minimal call

The stable model ID is gemini-4-neon. Migrating from another Gemini 4 model means changing only the model string; trackers report existing integrations switch without schema changes, consistent with the official model page. Entry points: the Gemini API paid tier (AI Studio, Android Studio), via generateContent and the Interactions API; a billing account with prepaid credit is required — there is no free tier, and a free API key will fail on Neon.

Model: gemini-4-neon thinking_level: low | medium | high (default) | xhigh max_output_tokens: 65,536 (sync default); long decode up to 1,048,576 Context window: 1,048,576 tokens (unified input + output) Input: text, image, audio, video, PDF Output: text (no image generation, audio generation or Live API)

On capabilities, the model page confirms: Computer Use, code execution, file search, grounding with Google Search / Maps, function calling, structured outputs, URL context and agentic video understanding; image generation, audio generation and the Live API are not supported. Batch API, Flex inference and Priority inference are all supported.

Gemini app: picker and subscriptions

Google AI Pro and Ultra subscribers pick Neon from the model dropdown in the Gemini app. 9to5Google reports the refreshed picker has four entries: Auto, Everyday help (4 Hydrogen), Deeper analysis (4 Neon), Code & complex work (4 Carbon), with Neon taking Argon's place and 3.1 Pro going away. A reminder: Google's official blog does not mention the picker change, so this is attributed to 9to5Google's reporting. In the app, per Google's blog, Neon suits work that needs frontier-level reasoning with fast responses — reading spreadsheets, PDFs and photos more precisely, turning dense material into reports, combining Gmail and Docs information, comparing prices across vendors online and assembling briefs with the latest data, and handing multi-step to-dos to Tasks in Gemini (free in some regions, but running them on Neon still needs Pro / Ultra).

Antigravity: the Orchestrator

Paid Pro / Ultra subscribers use Neon in the Google Antigravity desktop app, CLI and IDE extensions. The official Antigravity blog positions it as the sweet spot for "demanding daily workflows": repo-level edits, debugging, refactoring, secure code generation, grounded enterprise analysis and STEM reasoning, with reliable multi-step instruction following and the lowest fabrication rate yet in internal testing. Neon can act as an Orchestrator directing parallel Hydrogen subagents; Antigravity CLI 1.5.11 added the gemini-pro-orchestrator-neon variant. One caution: SWE-sweep scores 39.0% inside the Antigravity harness but only 12.8% on the third-party mini-SWE harness — the harness matters enormously; see our benchmarks article.

Enterprise: Gemini for business and Vertex AI

The enterprise route is Gemini for business (API or model dropdown). Vertex AI needs its own note: trackers report Neon is already listed in the Vertex AI Model Garden, Google Cloud's billing catalog and quota pages, and answers on Vertex's global endpoint; but Google's official blog only says "Gemini for business" without naming Vertex AI separately. This site writes the Vertex AI listing as "tracker-reported" until Google's own documentation confirms the wording. Enterprise users can also apply for the Fairwind Program (Neon for cyber) through their Cloud account team.

Long decode, caching and ignored parameters: the pitfalls list

ItemRuleWatch out
Max output65,536 tokens synchronous default; long decode up to 1,048,576The output cap covers thinking tokens — set it too low and you get an empty answer (the trap already confirmed on Carbon)
Thinking tierslow / medium / high (default) / xhighHigh is the best value (AA 60); open xhigh only for the hardest jobs (61, at 2.3x the tokens)
Context cachingImplicit only; auto-triggers at prompts of 4,096+ tokens, 95% off ($0.15)Explicit caching is not supported
Ignored parameterstemperature, topP, topK are ignored (tracker-reported)Not verified against official documentation in this round; do not use them to control cost
TimeoutsAA measured 21.5 min per index task at xhigh, 56.51s to first answer tokenAllow generous HTTP timeouts on long jobs; do not expect Hydrogen-like speed from Neon
Free tierNoneNeither free API keys nor free subscriptions reach Neon

Neon for cyber: Fairwind application only

The cybersecurity variant, Gemini 4 Neon for cyber, rolls out to Fairwind Program trusted defenders in the coming days, by application through a Cloud account team. Google-reported scores: AutoBaxBench hard tier 76% (beats Carbon across all tiers), SEC-bench Pro 90.5%, WebExploitBench v1.2 51.3%, CWE-bench v2 70.7% (3.4 points above Carbon), ExploitGym at Carbon's level with half the tokens. All are Google-reported, with no independent reproduction yet.

Our take: the fastest way onto Neon is the paid API: one model ID, four thinking tiers, long decode up to 1M tokens. Everyday users need two facts: the app needs a Pro subscription at minimum, and "Deeper analysis" in the picker is Neon. Developers need three: no free tier, implicit caching only, and do not set the output cap too low. Treat Vertex AI and the ignored-parameters list as pending until the official documentation confirms them.

Reference sources: this article is an original rewrite produced by this site; facts and figures were checked against Gemini API model page (read Nov 19, 2026), Gemini API pricing (read Nov 19, 2026), Google Antigravity blog (Nov 18, 2026) and Google's official blog (Nov 18, 2026)

← Back to the Gemini 4 Neon topic page