{"schemaVersion":"1.0","type":"TechArticle","types":["Article","TechArticle"],"slug":"gemini-4-neon-google-s-fast-frontier-level-model-for-everyday-heavy-lifting-1i7dk","url":"https://zyvop.com/gemini-4-neon-google-s-fast-frontier-level-model-for-everyday-heavy-lifting-1i7dk","title":"Gemini 4 Neon: Google's Fast, Frontier-Level Model for Everyday Heavy Lifting","subtitle":"Google’s new Gemini 4 Neon offers Carbon-level quality at much higher speed and a lower price. Here’s how it stacks up against Carbon, Claude and GPT-6.2.","tldr":"Google’s Gemini 4 Neon launched on November 18 with Carbon-level intelligence, roughly twice Carbon’s output speed and a 25% lower standard price. We break down the benchmarks, pricing and availability, and where it trails Claude Fable 5.2 and GPT-6.2 Astra.","keywords":["Gemini 4 Neon","benchmarks","google-ai","AI models","LLM Comparison","AI Models"],"entities":["ZyVOP","Gemini 4 Neon","benchmarks","google-ai","AI models","LLM Comparison","AI Models"],"keyTakeaways":["Gemini 4 Neon: Google's Fast, Frontier-Level Model for Everyday Heavy Lifting Published November 19, 2026 Google has been shipping models at a pace that is hard to keep up with.","On November 18, it added Gemini 4 Neon to the lineup, the fourth Gemini 4 model in under two months.","The pitch is simple: quality close to Google's top-tier model, at much higher speed and a lower price."],"headings":["Where Neon fits in the Gemini 4 family","What makes Neon interesting","1. Carbon-class results at Flash-like speed","2. Strong showing on coding agents","3. Better at visual and document work","4. Four effort levels","5. Long context and multimodal input","Pricing","Where you can use it","How good is it, really?","How Neon compares","Neon vs. Carbon","Neon vs. rival models","Who should care","The bigger picture","Further reading"],"outboundLinks":["https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/","https://9to5google.com/2026/11/10/gemini-app-4-carbon/","https://9to5google.com/2026/11/12/gemini-4-hydrogen-launch/","https://9to5google.com/2026/11/18/gemini-4-neon-announcement/","https://artificialanalysis.ai/articles/gemini-4-neon-intelligence-gain","https://www.androidpure.com/gemini-4-neon-gemini-app/","https://artificialanalysis.ai/articles/openai-has-the-2-model-in-the-intelligence-index-with-the-release-of-gpt-6-2-astra-and-gpt-6-2-sol","https://artificialanalysis.ai/articles/google-deepmind-multimodal-knowledge-agent","https://the-decoder.com/google-ships-gemini-4-neon-its-fourth-new-gemini-model-in-a-month/","https://mlq.ai/news/google-releases-gemini-4-neon-at-3-per-million-input-tokens-its-fourth-new-model-in-a-month/"],"contentText":"Gemini 4 Neon: Google's Fast, Frontier-Level Model for Everyday Heavy Lifting Published November 19, 2026 Google has been shipping models at a pace that is hard to keep up with. On November 18, it added Gemini 4 Neon to the lineup, the fourth Gemini 4 model in under two months. The pitch is simple: quality close to Google's top-tier model, at much higher speed and a lower price. This post covers what Neon is, how it fits into the Gemini 4 family, what it costs, and what the early benchmarks say. Where Neon fits in the Gemini 4 family Neon arrives right after Carbon and Hydrogen, which launched the week before. Here is how the family looks now: Model Role Hydrogen Everyday help, the replacement for Flash Neon Deeper analysis, coding and agentic work at high speed Carbon The most complex tasks, aimed at the limits of model intelligence Neon also takes over from Argon, the model that was previously positioned for code and analysis. In the Gemini app, the model picker now offers Auto, Everyday help (Hydrogen), Deeper analysis (Neon), and Code &amp; complex work (Carbon). What makes Neon interesting 1. Carbon-class results at Flash-like speed According to 9to5Google's launch report, Neon performs comparably to Carbon across coding, reasoning, and agentic tasks while running much faster and costing less. Artificial Analysis, an independent benchmarking firm, measured Neon at roughly 110 output tokens per second, versus about 57 for Carbon. Its time per task came out around 40% lower. 2. Strong showing on coding agents On the Artificial Analysis Coding Agent Index, Neon (at its highest effort level) scored 59 when used through Google's Antigravity CLI. Google's own announcement cites 58.6 on the same index and claims first place. That is ahead of Carbon's 55 and in the same range as Claude Fable 5.2 and Claude Opus 5.5 in Claude Code, which scored 58 each. The cost per task was reported as lower for Neon than for either of those Claude setups. 3. Better at visual and document work Artificial Analysis also reported notable gains in visual reasoning and output presentation quality. On its presentation-quality ranking, Neon took the top spot, ahead of the previous leader. That matters for people who use AI to produce slide decks, reports, and other polished documents. 4. Four effort levels Neon offers low, medium, high, and xhigh thinking levels. The xhigh setting was introduced with Hydrogen. This lets developers trade cost and latency against reasoning depth on a per-task basis. 5. Long context and multimodal input Neon supports a 1 million token context window and accepts text, images, video, audio, and PDFs as input, with text-only output. It also supports Computer Use, code execution, file search, and grounding through Google Search and Google Maps. Pricing Input (per 1M tokens) Output (per 1M tokens) Gemini 4 Neon $3 $15 Hydrogen about half of Neon's price about half of Neon's price Carbon (promo, until Nov 30) $2 $10 Carbon (standard) $4 $20 Two things stand out. First, Neon costs about 25% less than Carbon's standard rate, even though Carbon is currently discounted below Neon until the end of November. Second, Neon used roughly 15% more output tokens per task in Artificial Analysis's testing, so its savings come from lower token prices and faster output rather than from being more frugal with tokens. In Artificial Analysis's testing, Neon at its highest (xhigh) effort level cost about $6.44 per task, around 17% less than Carbon at high effort ($7.76 at standard pricing). At Neon's default high effort, Artificial Analysis found it matches Carbon's score at less than half the cost per task, and Google's own launch post claims a 32% lower cost per task than Carbon at that setting. These come from different comparisons, so treat them as separate data points rather than one figure. Cached input tokens are billed at $0.15 per million, a 95% discount on the standard input rate. Where you can use it Gemini app: available to Google AI Pro and AI Ultra subscribers in the \"Deeper analysis\" slot, replacing Argon. Gemini 3.1 Pro is being removed from the app. As of November 19, Google hasn't spelled out per-plan usage limits or any free-tier access, according to Android Pure. Google Antigravity: available to paid AI Pro and Ultra subscribers. Developers: available on the paid tier of the Gemini API in Google AI Studio and Android Studio, under the model ID gemini-4-neon. Enterprise: available in Gemini for business (via the API or the model dropdown) and the Gemini Enterprise Agent Platform. If you are deploying outside the US, it is worth confirming regional endpoint availability and pricing with your Google Cloud account team before committing to production use. How good is it, really? On the Artificial Analysis Intelligence Index, Neon at xhigh scored 61, one point above Carbon at high effort. That placed it third overall at the time of the report, behind Anthropic's Claude Fable 5.2 (65) and OpenAI's GPT-6.2 Astra (62). A few caveats are worth keeping in mind: Some of the benchmark numbers in launch coverage were chosen and presented by Google, even when the evaluations themselves are run by third parties. Leaderboard gaps of a point or two are small and can shift with new model versions or evaluation updates. Real-world results depend on your workload. A model that tops a coding-agent index may still feel different on your codebase, your documents, or your prompts. How Neon compares Neon vs. Carbon Carbon is Google's own top-tier model, so this is the comparison that matters most when choosing within the Gemini 4 family. Figures are from Artificial Analysis unless noted. Neon is tested at its xhigh setting for the Intelligence Index and cost-per-task numbers, and Carbon at high. Metric Gemini 4 Neon Gemini 4 Carbon Intelligence Index 61 60 Coding Agent Index (Antigravity CLI) 59 55 Cost per task $6.44 $7.76 (standard pricing) Cost per task when both score 60 (Neon at high, Carbon at high) $3.74 $7.76 (standard pricing) Time per task ~21 minutes ~36 minutes Output speed ~110 tokens/sec ~57 tokens/sec Visual reasoning (GDP.pdf, at high) 33.4% 28.4% Multimodal Knowledge Agent 56% 54% List price per 1M tokens (in / out) $3 / $15 $4 / $20 ($2 / $10 until Nov 30) Hydrogen, the cheaper everyday model, costs about half of what Neon does. One summary of Artificial Analysis data puts it at 51 on the Intelligence Index at its default high setting (54 at xhigh), well below Neon and Carbon. I could not find a full set of matching benchmark figures for it, so it is left out of the table. Neon vs. rival models Model Intelligence Index Cost per Intelligence Index task Coding Agent Index Coding cost per task Claude Fable 5.2 65 (1st, max setting) Not reported 58 (xhigh, in Claude Code) $15.47 GPT-6.2 Astra 62 (2nd, max setting) $4.00 Not reported Not reported Gemini 4 Neon 61 (3rd, xhigh) $6.44 59 (xhigh, in Antigravity CLI) $6.86 Claude Opus 5.5 Not reported Not reported 58 (max, in Claude Code) $12.61 \"Not reported\" means the figure was not in the sources I checked, not that it doesn't exist. A few things stand out: Neon is not the overall leader. It sits third on general intelligence, four points behind Claude Fable 5.2 and one behind GPT-6.2 Astra. GPT-6.2 Astra is cheaper per task. Artificial Analysis puts Astra's cost per Intelligence Index task at $4.00, so Neon's $6.44 is about 1.6 times higher, for a score one point lower. On coding agents it is level with the best, at lower cost. Its score is within a point of the two Claude setups, and its cost per task is roughly half of theirs. Presentation quality is a strength, but only one part of the picture. On Artificial Analysis's AA-Briefcase evaluation, Neon leads the presentation-quality category with 1771 Elo, ahead of the previous leader, Claude Sonnet 5.5, at 1726. On the overall AA-Briefcase score, though, Neon's 1700 trails Claude Fable 5.2 (1785) and Claude Sonnet 5.5 (1815). Multimodal knowledge work is close to a tie. On the Multimodal Knowledge Agent benchmark, Neon scores 56% against 57% for Claude Fable 5.2. One caution: these rows use different effort settings and different coding harnesses (Antigravity CLI for Neon, Claude Code for the Claude models), so the comparison is useful but not perfectly like-for-like. I left out rivals' list prices on purpose, because cost per task reflects real usage better than a per-token rate. Who should care Developers who want near-frontier coding and agent performance without frontier-level latency and cost. Teams building agents that make many calls per task, where speed and price compound quickly. Knowledge workers producing visual, document-heavy output such as decks and reports. Everyday Gemini subscribers, who get a noticeable upgrade in the \"Deeper analysis\" slot of the model picker. The bigger picture The most telling detail may be the cadence. Four Gemini 4 models in under two months suggests Google is competing hard on speed of iteration as well as raw capability, and that the gap between the top few labs is narrow enough that price and latency now matter as much as the leaderboard. Neon is not trying to be the single smartest model Google offers. It is trying to be the one you will actually use most of the time. For a lot of coding and analysis work, that is a good bet. Further reading Introducing Gemini 4 Neon (Google, official announcement) Google continues to accelerate its release cadence with Gemini 4 Neon (Artificial Analysis) Google launches Gemini 4 Neon with Carbon quality at Flash speeds (9to5Google) Google ships Gemini 4 Neon, its fourth new Gemini model in a month (The Decoder) Google releases Gemini 4 Neon at $3 per million input tokens (MLQ) Gemini 4 Neon reaches the Gemini app for AI Pro and Ultra users (Android Pure) Figures reflect coverage from November 18–19, 2026, and may change as pricing promotions end and new evaluations are published.","contentHash":"sha256:9e00ecf045871f2e2a2c62531fff2f4f04f9efc49dfd5de51400f32c59c4546a","authorName":"ZyVOP","authorUrl":"https://zyvop.com/author/zyvop","authorSameAs":["https://zyvop.com","https://github.com/zyvop","https://x.com/zyvop1"],"category":"AI Models","tags":["Gemini 4 Neon","benchmarks","google-ai","AI models","LLM Comparison"],"audience":"Software engineers and developers building applications with AI Models","tone":"Instructional, practical, code-first","readingTimeMinutes":7,"wordCount":1693,"faqs":null,"primaryTopic":"AI Models","publishedAt":"2026-11-19T05:41:51.533Z","updatedAt":"2026-11-19T05:41:51.533Z","canonicalUrl":"https://zyvop.com/gemini-4-neon-google-s-fast-frontier-level-model-for-everyday-heavy-lifting-1i7dk"}