Google released Gemini 4 Neon on November 18, 2026, at $3 input, $0.15 cached input, and $15 output per million tokens, sitting between Gemini 4 Hydrogen and Gemini 4 Carbon. The model (gemini-4-neon) is available on the paid Gemini API tier, in Google Antigravity, and in the Gemini app for Pro and Ultra subscribers.
What is Gemini 4 Neon and when was it released?
Gemini 4 Neon is Google’s mid-tier Gemini 4 reasoning model, introduced on November 18, 2026, for complex software engineering, knowledge work, and science workflows. The launch was announced on the official Google blog by Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect at Google, and the Gemini API deprecations table records November 18, 2026, as the release date for gemini-4-neon with no shutdown date announced.
Google positions Gemini 4 Neon specifically between its two existing Gemini 4 releases. In the announcement blog, Google states: “While Carbon tackles your hardest, long-horizon projects and Hydrogen handles fast everyday execution, Neon balances depth and efficiency. Neon delivers frontier-level reasoning for complex, high-frequency workflows faster and more cost efficiently.”
Google’s developer release notes add practical guardrails around that split. Teams should use Neon for complex software engineering, knowledge work, and science workflows where fast end-to-end completion matters, reserve Carbon for the hardest architecture, debugging, science, and security work, keep Hydrogen as the default for everyday coding and agentic workflows, and keep Gemini 3.5 Flash-Lite for high-volume, low-latency tasks. Inside Google Antigravity, Neon can run standalone or act as what Google calls a “fast, high-judgment Orchestrator directing parallel Hydrogen subagents,” and the Antigravity agent in Gemini Managed Agents supports gemini-4-neon in both configurations.
How much does the Gemini 4 Neon API cost?
The Gemini 4 Neon API (gemini-4-neon) costs $3.00 per million input tokens, $0.15 per million cached input tokens, and $15.00 per million output tokens on the paid tier as of November 19, 2026. Pricing stays flat across the full unified context window, with no tiered step-up for longer prompts.
Output billing includes all thinking tokens generated during reasoning, plus internal tool system instructions and function declarations when grounding tools are enabled. Cached input tokens receive a 95% discount over the standard $3.00 input rate through implicit caching, which runs automatically on prompts of at least 4,096 tokens. Explicit context caching via cachedContents or caches.create is not supported on Neon, so developers migrating from older Gemini endpoints need to remove cached_content request fields.
Gemini 4 Neon has no free API tier. On the paid tier, Google’s pricing documentation marks “Used to improve our products” as No. If your application uses grounding on the paid tier, Grounding with Google Search includes 5,000 free search requests per month shared across Gemini 3 and newer models, then $14 per 1,000 requests, subject to a daily cap of 1,500 requests per day, while Grounding with Google Maps shares a 5,000-request monthly free allowance before billing $14 per 1,000 search queries.
Technical specs and thinking levels
Gemini 4 Neon accepts text, image, video, audio, and PDF inputs and produces text output within a 1,048,576-token unified context window shared across input, thinking, and output tokens. Default synchronous output caps at 65,536 tokens, while Google’s long-decode guide allows up to 1,048,576 minus input tokens via background execution or client-side continuation.
The model runs on both generateContent and the Interactions API, and lists support for Batch API, Flex inference, and Priority inference (though Google’s pricing table publishes only standard paid-tier rates as of November 19, 2026). Supported capabilities include thinking, structured outputs, code execution, computer use, function calling, file search, URL context, Search and Maps grounding, agentic video understanding, and implicit caching. Image generation, audio generation, and the Live API are unsupported.
Developers can configure four thinking levels: low, medium, high (the default), and xhigh. Passing minimal returns an error. If you are migrating existing calls, Google also advises removing custom temperature, top_p, and top_k parameters because in-range values are ignored and out-of-range values trigger a 400 validation error.
Where can you use Gemini 4 Neon today?
Gemini 4 Neon is rolling out across Google’s paid developer endpoints, enterprise workspace tools, and paid consumer tiers, with no free API tier or unpaid Gemini app access. Google’s November 18 announcement and developer docs list five launch surfaces:
- Gemini API paid tier: Available via
gemini-4-neonin Google AI Studio and Android Studio acrossgenerateContentand the Interactions API. - Gemini for business: Rolling out through the API and the model dropdown for business customers.
- Google Antigravity: Available to paid Google AI Pro and Ultra subscribers (Pro and Ultra subscribers can also try Neon directly in Google AI Studio).
- Gemini app: Rolling out in the model drop-down for Pro and Ultra subscribers only.
- Fairwind Program (cybersecurity): A specialized “Gemini 4 Neon for cyber” release is rolling out over the coming days to eligible defenders, who can request access through their Google Cloud account team.
How does Gemini 4 Neon compare with Carbon and Hydrogen?
Within the Gemini 4 family, Gemini 4 Neon sits between Gemini 4 Hydrogen and Gemini 4 Carbon in task complexity, while its per-token list price sits temporarily above Carbon’s introductory rate through November 30, 2026.
Checking the official Gemini Developer API pricing page on November 19, 2026, reveals an important calendar detail:
- Gemini 4 Hydrogen (
gemini-4-hydrogen): $1.50 input, $0.075 cached input, and $7.50 output per million tokens, plus a rate-limited free API tier (where free-tier data may be used to improve Google products). - Gemini 4 Neon (
gemini-4-neon): $3.00 input, $0.15 cached input, and $15.00 output per million tokens, paid tier only, with no scheduled December price change listed. - Gemini 4 Carbon (
gemini-4-carbon): $2.00 input, $0.10 cached input, and $10.00 output per million tokens through November 30, 2026, increasing to $4.00 input, $0.20 cached input, and $20.00 output starting December 1, 2026 (matching Gemini 4 Argon’s schedule).
Through November 30, 2026, Carbon’s introductory per-token list price is lower than Neon’s; starting December 1, 2026, Carbon becomes more expensive per token than Neon. Official positioning separates the two models by task scope and reasoning volume rather than token rate alone. Google’s announcement blog states that at its default high thinking setting, Neon delivers a “32% lower cost per task than Carbon” because total spend depends on how many thinking and output tokens each model generates to finish a job. Note also that Neon and Hydrogen default to high thinking and support xhigh, whereas Carbon and Argon default to medium and cap at high.
Gemini 4 Neon vs. GPT-6.2 Sol and Claude on list price
On published November 2026 list prices, Gemini 4 Neon at $3 input and $15 output per million tokens sits between the $2/$10 mid-tier (GPT-6.2 Sol and Claude Sonnet 5.5) and the $4/$20 tier (Claude Opus 5.5 and post-December Gemini 4 Carbon).
The table below compiles Google’s official paid-tier rates checked November 19, 2026, alongside OpenAI and Anthropic list rates documented in Lotook’s published November 2026 AI API pricing comparison and GPT-6.2 coverage.
| Model | Input / 1M | Cached input / 1M | Output / 1M | Status as of November 19, 2026 |
|---|---|---|---|---|
| Gemini 4 Hydrogen | $1.50 | $0.075 | $7.50 | Paid tier; rate-limited free API tier available |
| GPT-6.2 Sol | $2.00 | $0.10 | $10.00 | Published Nov 17, 2026 OpenAI list price |
| Claude Sonnet 5.5 | $2.00 | $0.10 | $10.00 | Published Nov 2026 Anthropic list price |
| Gemini 4 Carbon (until Nov 30) | $2.00 | $0.10 | $10.00 | Intro paid-tier rate through Nov 30, 2026 |
| Gemini 4 Neon | $3.00 | $0.15 | $15.00 | Paid API only; output includes thinking tokens |
| Gemini 4 Carbon (from Dec 1) | $4.00 | $0.20 | $20.00 | Scheduled paid-tier rate starting Dec 1, 2026 |
| Claude Opus 5.5 | $4.00 | $0.20 | $20.00 | Published Nov 2026 Anthropic list price |
| GPT-6.2 Astra | $8.00 | $0.40 | $40.00 | Published Nov 17, 2026 OpenAI list price |
| Claude Fable 5.2 | $8.00 | Not listed | $40.00 | Published Nov 2026 Anthropic list price |
Per-token list rates do not directly predict monthly API bills, because reasoning models bill thinking tokens as output. A model priced at $3/$15 can cost less or more per completed task than a $2/$10 or $4/$20 peer depending on prompt structure, implicit cache hit rate, and thinking depth.
Official evaluation claims from Google’s announcement
Google’s November 18, 2026 announcement restates several third-party benchmark results and internal cost-per-task estimates for Gemini 4 Neon, none of which Lotook has independently tested or verified. Selected figures cited by Google include:
- Artificial Analysis Intelligence Index: Google reports Neon ranks #3 overall with a score of 61 at
xhighthinking. - Vals Index and domain benchmarks: Google cites a #2 rank on the Vals Index (70.9%) and #1 positions on Legal Research Bench and Tax Agent Bench.
- Agent Arena Index: Google states Neon debuts at #2 globally (score 1336 across 4,200+ real-world agent sessions) for mid-task steerability.
- Coding and terminal evaluations: Google reports #1 overall on the Artificial Analysis Coding Agent Index (58.6), 65.2% on Terminal-Bench 4.1, and 46.2% on FrontierCode 2.0 at default
highthinking. - SWE-sweep harness distinction: Google reports Neon reaches 12.8% on SWE-sweep under the minimal mini-SWE harness and “more than triples to a state-of-the-art 39.0% in Google Antigravity.” That spread reflects the impact of the Antigravity agent harness alongside the underlying model.
- Science and agentic tasks: Google cites 70.5% on Terminal-Bench Science, 84.0% on RiemannBench, 71.3% on HLE-Diamond, 56.3% on Zapier’s AutomationBench (#1), and $16,843 on Andon Labs’ Vending-Bench 2 (#1).
On efficiency, Google claims Neon achieves 50% to 80% lower cost per task than competing frontier models on the Vals Index and Agent Arena at high, and over 36% lower cost per task at xhigh relative to competing models at their highest setting. Because these are vendor-reported percentages without published dollar-per-task logs, teams should benchmark on their own prompts before switching production traffic.
Choosing between Gemini 4 Neon, Carbon, Hydrogen, and peers
Choosing among Gemini 4 Neon, its Gemini 4 siblings, and competing models comes down to task complexity, budget structure, and platform fit rather than a single universal winner. Our broader guide on how to choose an AI model covers workload evaluation in more detail, and the practical split for November 2026 looks like this:
- Choose Gemini 4 Hydrogen, GPT-6.2 Sol, or Claude Sonnet 5.5 for everyday coding, high-volume pipelines, or lower-cost $1.50/$7.50 and $2/$10 workloads, or when you need Hydrogen’s rate-limited free API tier for prototyping.
- Choose Gemini 4 Neon when your team runs complex software engineering, knowledge work, or science workflows on the paid Gemini API, uses Gemini Pro/Ultra, or wants an orchestrator paired with Hydrogen subagents in Google Antigravity.
- Choose Gemini 4 Carbon for long-horizon architecture, security, and debugging problems that demand maximum depth, keeping in mind Carbon’s $2/$10 introductory rate ends November 30, 2026, and rises to $4/$20 on December 1.
- Choose Claude Opus 5.5 or $8/$40 flagships (GPT-6.2 Astra, Claude Fable 5.2) if your production stack is already built around Anthropic or OpenAI SDKs and needs higher-tier reasoning without migrating providers.
FAQ
Is Gemini 4 Neon free to use on the Gemini API or Gemini app?
No, Gemini 4 Neon has no free API tier and is not available to unpaid Gemini app users. Developer access requires the paid tier of the Gemini API in Google AI Studio or Android Studio. Consumer and workspace access requires a paid Google AI Pro or Ultra subscription in the Gemini app or Google Antigravity, or Gemini for business.
Why is Gemini 4 Carbon cheaper per token than Neon right now?
Gemini 4 Carbon carries an introductory paid-tier list price of $2 input and $10 output per million tokens through November 30, 2026, making its per-token rate lower than Gemini 4 Neon’s $3 and $15 rates during that window. Starting December 1, 2026, Carbon moves to $4 input and $20 output per million tokens, putting Neon below Carbon on standard list rates.
Does Gemini 4 Neon support explicit context caching?
No, Gemini 4 Neon supports implicit context caching only. Google’s documentation notes that implicit caching runs automatically on prompts of at least 4,096 tokens and applies a 95% discount ($0.15 per million tokens) on cache hits. Explicit caching APIs such as cachedContents and caches.create are unsupported, so developers migrating from older Gemini models must remove cached_content fields.
Which thinking levels does Gemini 4 Neon support?
Gemini 4 Neon supports four thinking levels: low, medium, high (the default), and xhigh. Passing minimal returns an error. Because output billing at $15 per million tokens includes thinking tokens, choosing low or medium reduces billed output on simpler prompts, while xhigh reserves deeper reasoning for harder requests. Carbon and Argon default to medium and cap at high.
What is the difference between Neon’s 12.8% and 39.0% SWE-sweep scores?
According to Google’s November 18 announcement, 12.8% is Gemini 4 Neon’s score on SWE-sweep when evaluated with the minimal mini-SWE harness, whereas 39.0% is the score when running inside Google Antigravity. The jump shows the impact of Google’s Antigravity agent harness rather than a raw model-only change, and Lotook has not independently verified either figure.
Sources
- Introducing Gemini 4 Neon - Google Official Blog (November 18, 2026)
- Gemini Developer API Pricing - Google AI Studio
- Gemini 4 Neon Model Documentation - Google AI Studio
- What’s New in Gemini 4 Neon - Google AI Studio
- Long Decode Guide - Google AI Studio
- Gemini API Deprecations and Release Dates - Google AI Studio
- @GoogleDeepMind Launch Announcement on X (November 18, 2026)



