Google launches Gemini 4 Neon at $3/$15, 25% below Carbon's December rate

Google has released Gemini 4 Neon, a new member of the Gemini 4 family that sits between the top-end Gemini 4 Carbon and the fast, inexpensive Gemini 4 Hydrogen. The model was announced on November 18, 2026 in a post by Koray Kavukcuoglu, SVP of Google DeepMind and Chief AI Architect at Google, and it went live the same day on the paid tier of the Gemini API, in Google AI Studio, in the Gemini app for Google AI Pro and Ultra subscribers, in Google Antigravity and in Gemini for business.
The developer price is $3 per 1M input tokens and $15 per 1M output tokens, with cached input at $0.15 per 1M tokens, and the rate stays flat across the full 1,048,576-token context window. That places Neon above Carbon's current introductory price of $2/$10, but from December 1, 2026 Carbon and Argon move to $4/$20 under Google's published schedule, so for most of its life Neon will cost 25% less per token than Carbon on both input and output.
Google describes Neon as a model that brings frontier-level intelligence to fast, end-to-end task completion in software engineering, knowledge work and science. According to the model card, it is adapted from Gemini 4 Carbon, accepts text, images, audio and video, produces text, and has a knowledge cutoff of March 2026. It defaults to the high thinking level and adds an xhigh level above it, a setting Carbon and Argon do not offer.
A security-focused variant is also planned. Google says trusted defenders enrolled in its Fairwind Program will start getting access to a cyber version of Neon over the next few days, and the model card lists cyber defense, including secure code generation, penetration testing and vulnerability patching, among the intended uses.
··········
GEMINI 4 NEON API PRICING COMPARED WITH CARBON, HYDROGEN AND RIVALS.
Google lists Neon's prices in its launch post and on the Gemini API "What's new in Gemini 4 Neon" page. Output pricing includes thinking tokens, which is relevant for a model that defaults to high reasoning effort. Cached input applies through implicit caching only, with a 95% discount on cache hits for prompts of at least 4,096 tokens. Explicit context caching through cachedContents is not supported, so there is no separate cache storage fee to budget for.
Google has not published Batch, Flex or Priority prices for Neon at launch, and the general Gemini API pricing page did not yet list the model when we checked it. Developers who rely on Batch discounts for Carbon, Argon or other Gemini models should treat Neon as a Standard-tier model until Google updates that page.
The table below sets Neon against the other Gemini 4 models and against the closest models from Anthropic and OpenAI. Prices as listed on the official pricing pages of Google, Anthropic and OpenAI on November 18, 2026.
........
Model | Input | Cached input | Output | Batch input / output | Effective date |
Gemini 4 Neon | $3.00 | $0.15 | $15.00 | Not published | Nov 18, 2026 |
Gemini 4 Carbon (current) | $2.00 | $0.10 | $10.00 | Not listed on Carbon page | Until Nov 30, 2026 |
Gemini 4 Carbon (scheduled) | $4.00 | $0.20 | $20.00 | Not listed on Carbon page | From Dec 1, 2026 |
Gemini 4 Argon (scheduled) | $4.00 | $0.20 | $20.00 | $2.00 / $10.00 | From Dec 1, 2026 |
Gemini 4 Hydrogen | $1.50 | $0.075 | $7.50 | Not listed on Hydrogen page | Current |
Claude Opus 5.5 | $4.00 | $0.20 (cache hit) | $20.00 | $2.00 / $10.00 | Current |
GPT-6.2 Sol | $2.00 | $0.10 | $10.00 | $1.00 / $5.00 | Current |
........
All figures are USD per 1M tokens. Claude Opus 5.5 also charges for cache writes ($5 per 1M tokens for a five-minute cache and $8 for a one-hour cache), while Gemini and OpenAI apply caching automatically without a write fee. Argon's Standard price follows the same schedule as Carbon, $2/$10 until November 30 and $4/$20 from December 1.
The pricing positions Neon in a clear slot. It costs twice as much as Hydrogen per token and undercuts Claude Opus 5.5 and post-December Carbon by a quarter. GPT-6.2 Sol remains cheaper per token at $2/$10, so any case for Neon against Sol rests on task results and token efficiency rather than list price.
··········
WHAT A CODING AGENT AND A SUPPORT CHATBOT COST ON GEMINI 4 NEON.
Per-token prices only tell part of the story, so we ran two typical workloads through the official rates. Both examples assume identical token counts for every model, which is a simplification: models that think longer or retry more often will consume more tokens, and Google's own claim is that Neon finishes tasks with fewer tokens than Carbon. The figures show the price effect alone.
The first workload is a coding agent session of 50 model calls, each carrying 80,000 input tokens of repository context and history, with 90% of that input served from cache, and 2,500 output tokens per call including thinking. That totals 400,000 fresh input tokens, 3.6M cached tokens and 125,000 output tokens per session.
The second workload is a customer-support chatbot handling 10,000 conversations a day for 30 days. Each conversation sends 6,000 input tokens, 75% of which is a stable system prompt and knowledge snippet long enough to qualify for implicit caching, and returns 500 output tokens. That is 450M fresh input tokens, 1.35 billion cached tokens and 150M output tokens a month.
........
Model | Coding agent session | Support chatbot, per month |
Gemini 4 Neon | $3.61 | $3,802.50 |
Gemini 4 Carbon (until Nov 30) | $2.41 | $2,535.00 |
Gemini 4 Carbon (from Dec 1) | $4.82 | $5,070.00 |
Gemini 4 Hydrogen | $1.81 | $1,901.25 |
Claude Opus 5.5 (excluding cache writes) | $4.82 | $5,070.00 |
GPT-6.2 Sol | $2.41 | $2,535.00 |
........
For the agent session, Neon's cost breaks down into $1.20 for fresh input, $0.54 for cached input and $1.88 for output. Output is the largest single item, which is typical for reasoning models whose thinking tokens are billed as output. In the chatbot case, output again dominates at $2,250 of the $3,802.50 monthly bill on Neon.
A team that moved a Carbon workload to Neon today would pay about 50% more per token, because Carbon is still on its introductory rate. After December 1, the same switch lowers the bill by 25% at equal token counts, from $4.82 to $3.61 per agent session and from $5,070 to $3,802.50 a month for the chatbot. Google's separate claim of a 32% lower cost per task than Carbon at the high setting goes beyond this arithmetic, since it also counts the tokens each model uses to finish a task, and Google has not specified which Carbon price that comparison assumes.
The change favors teams running agents and long-context workloads that were planning to stay on Carbon or Argon after the price increase. High-volume chat applications with simple prompts remain cheaper on Hydrogen, which costs half as much as Neon and still has a free tier.
··········
GEMINI 4 NEON AVAILABILITY BY PLAN AND PLATFORM.
Neon is a paid model across every surface. Google states that it is available on the Gemini API paid tier only, through both generateContent and the Interactions API, and not on the free tier. Consumers reach it through Google AI Pro and Google AI Ultra.
........
Surface | Who gets Gemini 4 Neon | Status |
Gemini API | Paid-tier developers (generateContent and Interactions API) | Available Nov 18 |
Google AI Studio | Paid API tier; Google AI Pro and Ultra subscribers | Available Nov 18 |
Gemini app | Google AI Pro and Ultra subscribers | Available Nov 18 |
Google Antigravity | Google AI Pro and Ultra subscribers | Available Nov 18 |
Gemini for business | Via API and the model dropdown | Available Nov 18 |
Android Studio | Listed by Google among launch surfaces | Available Nov 18 |
Neon cyber variant | Trusted defenders in the Fairwind Program | Rolling out over the next few days |
........
Google's announcement does not give regional restrictions, rate limits or daily usage caps for Neon in the Gemini app. The API documentation refers developers to the general rate-limits page. For the plan prices and the limits each subscription carries, see our guide to Gemini models across the app, API and Vertex AI, and for the full developer price list our Gemini API pricing guide.
··········
GEMINI 4 NEON SPECIFICATIONS AND MIGRATION FROM CARBON OR HYDROGEN.
Neon shares the Gemini 4 platform conventions introduced with Argon, Carbon and Hydrogen. The context window is 1,048,576 tokens, shared by input, thinking and output. A synchronous response returns up to 65,536 tokens, which is also the default output limit, and longer outputs up to the full window require background execution with a higher max_output_tokens value.
........
Specification | Gemini 4 Neon | Gemini 4 Carbon | Gemini 4 Hydrogen |
Context window | 1,048,576 tokens | 1,048,576 tokens | 1,048,576 tokens |
Max synchronous output | 65,536 tokens | 65,536 tokens | 65,536 tokens |
Thinking levels | low, medium, high, xhigh | low, medium, high | low, medium, high, xhigh |
Default thinking level | high | medium | high |
Free API tier | No | No | Yes |
Cached input price | $0.15 | $0.10, then $0.20 | $0.075 |
........
Migration from Carbon, Argon or Hydrogen only requires changing the model ID to gemini-4-neon, with no request schema changes. Teams coming from Gemini 3 or earlier have more work to do. Google's migration notes say to remove low max_output_tokens caps or raise them to 65,536, set thinking_level explicitly if the high default changes behavior, delete cached_content fields, and remove temperature, top_p and top_k, which return a 400 error when out of range and are ignored otherwise. The minimal thinking level is rejected with an error.
Google also recommends raising HTTP client timeouts to at least 15 minutes or switching to background execution, and it reminds developers that billing covers every processed token, including thinking tokens, internal grounding instructions and function declarations. Supported features include function calling, structured outputs with JSON schemas, implicit caching, background execution, thought signatures for multi-turn state and Gemini Managed Agents, including the Antigravity agent. Google Search grounding is billed separately per query.
··········
GEMINI 4 NEON BENCHMARKS AND GOOGLE'S COST-PER-TASK CLAIMS.
Google published a long list of results with the launch. These are vendor-reported figures, and several come from third-party leaderboards whose rankings can change as new models arrive.
........
Benchmark | Gemini 4 Neon result (Google-reported) |
Artificial Analysis Intelligence Index (xhigh) | 61, ranked #3 |
Artificial Analysis Coding Agent Index | 58.6, ranked #1 |
Vals Index | 70.9%, ranked #2 |
Agent Arena Index | 1336, ranked #2 |
Terminal-Bench 4.1 | 65.2% |
Terminal-Bench Science | 70.5% |
Agents' Last Exam v1.1 | 75.8% |
Zapier AutomationBench | 56.3%, ranked #1 |
Vending-Bench 2 | $16,843, ranked #1 |
HLE-Diamond | 71.3% |
FrontierCode 2.0 (high) | 46.2% |
........
On cost, Google says that at the high setting Neon's cost per task is 32% lower than Carbon's and roughly 50% to 80% lower than top competing frontier models on the Vals and Agent Arena Pareto frontiers. At xhigh, Google puts the saving at 36% or more against competitors running at their highest settings. Google did not name the competing models in the announcement, so these comparisons cannot be checked model by model from the post.
The security results carry more weight here because of the planned cyber variant. Google reports 76% on the hard tier of AutoBaxBench, 51.26% on WebExploitBench v1.2, 90.5% on SEC-bench Pro and 70.7% on CWE-bench v2, which it describes as a 3.4% uplift over Carbon. On behavior, Google cites a 0.09% misaligned action rate on its agentic safety suite, a 99.2% pass rate on an agent honesty evaluation and shortcuts on 2.45% of tasks in a coding cheating evaluation, the lowest of any Gemini 4 model.
The model card adds context on risk. Under Google's Frontier Safety Framework, Neon showed a general capability increase over Argon that stayed below the threshold requiring a full new set of critical capability evaluations, so Argon's determinations were carried forward. Google states that the critical capability level for biological uplift and for cybersecurity uplift cannot be ruled out, and that mitigations for those areas were strengthened. The card lists hallucinations and jailbreak resistance as known limitations, and says the full Safety and Security Report is still to come.
··········
GEMINI 4 NEON: CONFIRMED FACTS AND OPEN QUESTIONS.
The confirmed facts are concrete. Gemini 4 Neon is live on the paid Gemini API at $3 per 1M input tokens, $0.15 cached and $15 output, with a flat price across a 1,048,576-token window, 65,536 tokens of synchronous output and four thinking levels defaulting to high. Consumers get it through Google AI Pro and Ultra in the Gemini app, AI Studio and Antigravity, and developers on Carbon, Argon or Hydrogen can switch by changing a single model ID.
The open points concern cost and access details that Google has not yet published. Batch, Flex and Priority rates for Neon are missing, and so are rate limits and app usage caps for each subscription. The cost-per-task advantage over Carbon and competing models rests on Google's own measurements, with the competitors unnamed and the Carbon price basis unstated. For buyers, the decisive date is December 1: once Carbon and Argon double to $4/$20, Neon becomes the cheaper way to run Google's stronger reasoning models, while Hydrogen and GPT-6.2 Sol remain the lower-cost options for workloads that do not need Neon's extra depth.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org




