Quick answer
Gemini 4 Neon is Google’s new mid-priced Gemini 4 model, released on 18 November 2026, and the early tests suggest it gets close to the best models from Anthropic and OpenAI while running faster and costing less per task. It costs $3 per million input tokens and $15 per million output tokens, has a 1 million-token context window and four thinking levels. In WorldofAI’s hands-on review it shone at SVG graphics, 3D scenes and front-end design, but premium models such as Claude Fable 5.2 and Opus 5.5 still produced slightly better results on some builds.
Key takeaways
- It is official. Google announced Neon on 18 November 2026. It is on the paid Gemini API, in Google AI Studio, in Google Antigravity and in the Gemini app for AI Pro and Ultra subscribers.
- The headline is value, not raw power. Google says Neon ranks #3 on the Artificial Analysis Intelligence Index (score 61 at its highest thinking level) and #2 on the Vals Index.
- It is fast. Figures quoted in the video put its output speed at roughly 110 tokens per second, nearly double Gemini 4 Carbon.
- Hands-on results were strong but not flawless. SVGs and 3D scenes impressed; a shooter-game clone and a Minecraft-style build needed refinement.
- Test it on your own work before switching. Benchmarks and creator demos are a starting point, not proof it suits your tasks.
🛠️ Practical guides
Ready to put this into practice? These step-by-step guides answer two common follow-up questions:
- How to Try Gemini 4 Neon in Google AI Studio and Pick a Thinking Level — where to find Neon, what it costs to try and which of the four thinking levels to start with.
- How to Split a Project Between a Fast, Cheap AI Model and a Premium One — a simple workflow for drafting with a fast model and polishing with a premium one, so you pay top prices only where it matters.
In this guide
- Watch: the source video
- What is Gemini 4 Neon?
- How much does Gemini 4 Neon cost?
- What do the benchmarks say?
- How did Gemini 4 Neon do in the hands-on tests?
- Where does Gemini 4 Neon fall short?
- Is Neon really a Flash-class model?
- Should you use Gemini 4 Neon for your work?
- Frequently asked questions
- The bottom line
- Related posts
Every few weeks a new AI model claims to beat the rest. Gemini 4 Neon is more interesting than most because the pitch is not “smartest ever” but “nearly as good, much faster and cheaper”. For a small business or solo creator paying for AI by the task, that is the claim that actually matters.
This Gemini 4 Neon review is based on the hands-on test video from WorldofAI, with the facts checked against Google’s own announcement. We have kept the creator’s opinions separate from what Google has confirmed, and flagged anything that is still speculation.
Watch: the source video
What this video covers: WorldofAI’s full test of Gemini 4 Neon: the benchmark results, pricing, Artificial Analysis findings on speed and cost, and hands-on demos including a three.js battle scene against Claude Fable 5.2, a shooter-game clone built in Antigravity, SVG tests against GPT-6.2 Sol, Hydrogen and Opus 5.5, 3D scenes, an animated poem and a Minecraft-style game made in the Gemini app.
Video by WorldofAI (@intheworldofai). The video was sponsored by Mammouth, an AI platform that bundles several models from $10 a month. We have no link with the sponsor.
What is Gemini 4 Neon?
Gemini 4 Neon is a Gemini 4 model that sits between the fast, low-cost Gemini 4 Hydrogen and the larger Gemini 4 Carbon. Google’s developer notes describe it as built for “high-frequency complex tasks”: coding, knowledge work and science workflows where you want strong results without waiting ages or paying flagship prices.
| Detail | What Google says |
|---|---|
| Released | 18 November 2026 |
| Context window | 1 million tokens, with long outputs up to 1 million tokens |
| Thinking levels | Four: low, medium, high (default) and xhigh |
| Where you can use it | Paid Gemini API, Google AI Studio, Google Antigravity (AI Pro and Ultra), Gemini app (AI Pro and Ultra), Gemini for business |
| Free tier | Not available on the free Gemini API tier |
Google also says Neon has its lowest fabrication rate yet in internal Antigravity testing. Made-up answers have been a long-standing complaint about Gemini models, and the creator called this out as one of the biggest improvements. It is still worth checking for yourself; our guide How to Test an AI Model for Made-Up Answers Before You Rely on It shows how.
If you have been following the Gemini 4 family, see our earlier coverage of Gemini 4 Hydrogen and Gemini 4 Carbon.
How much does Gemini 4 Neon cost?
Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens, with cached input at $0.15 per million (a 95% discount). Thinking tokens count as output, so higher thinking levels cost more. Pricing is flat across the full 1 million-token window, according to Google’s pricing page.
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Notes |
|---|---|---|---|
| Gemini 4 Neon | $3.00 | $15.00 | Cached input $0.15 |
| Gemini 4 Carbon (offer) | $2.00 | $10.00 | Promotional price until 30 November |
| Gemini 4 Carbon (from 1 Dec) | $4.00 | $20.00 | Neon becomes 25% cheaper per token |
| Claude Opus 5.5 | $4.00 | $20.00 | Standard rates, as reported by VentureBeat |
Per-token prices only tell half the story. What you really pay is the cost per finished task, which depends on how many tokens a model burns to get there. Google says that at its default high setting Neon has a 32% lower cost per task than Carbon, and roughly 50% to 80% lower cost per task than top competing models on two independent leaderboards.
💡 Tip: Carbon is cheaper per token until 30 November. If you are running a big batch job this month, compare both on a small sample first. Our guide How to Compare AI Models on Token Cost, Not Just Output Quality walks you through it.
What do the benchmarks say?
Google and independent leaderboards put Neon just behind the biggest, slowest models, and ahead of most models in its price range. Here is what is confirmed by Google, and what was quoted in the video from Artificial Analysis.
| Claim | Source | Status |
|---|---|---|
| #3 overall on the Artificial Analysis Intelligence Index (61 at xhigh) | Google announcement | Confirmed by Google |
| #2 on the Vals Index (70.9%), on the cost/performance frontier | Google; VentureBeat | Confirmed by Google |
| #2 on steerability in the Agent Arena Index | Google announcement | Confirmed by Google |
| Tasks finished about 40% faster than Carbon (36 to 21 minutes) | Artificial Analysis, as quoted in the video | Reported |
| Output speed up from about 57 to about 110 tokens per second | Artificial Analysis, as quoted in the video | Reported |
| Coding Agent Index score of 59, level with Fable 5.2 and Opus 5.5 | Artificial Analysis, as quoted in the video | Reported |
| About $6.86 per coding task vs $12.61 (Opus 5.5) and $15.41 (Fable 5.2) | Artificial Analysis, as quoted in the video | Reported |
The creator also highlighted strong scores on finance, accounting, spreadsheet and automation benchmarks. Treat those as a sign that Neon is worth testing for office-style tasks, not as a promise it will handle your spreadsheets perfectly.
⚠️ Warning: Benchmarks measure narrow tasks under test conditions. A model that scores well can still struggle with your particular documents, prompts or tools, so always run your own small test.
How did Gemini 4 Neon do in the hands-on tests?
The hands-on results were impressive for the price, with the best work in SVG graphics, 3D scenes and front-end design. Here is a summary of each test from the video, with our plain-English take on what it means for you.
| Test | What happened | What it suggests |
|---|---|---|
| Medieval battle scene in three.js vs Claude Fable 5.2 | Fable 5.2 looked a little better; Neon finished faster at roughly half the cost | For many jobs the small quality gap may not justify double the price |
| Front-end web pages | Much better design taste than older Gemini models: scroll effects, typography, layouts | Worth trying for landing pages and mock-ups |
| Shooter-game clone in Antigravity (one prompt) | Detailed: mini-map, reloading, lighting, textures; some areas looked bland | A strong first draft; the creator felt Opus might have done better |
| SVG Xbox controller vs GPT-6.2 Sol and Gemini 4 Hydrogen | Neon produced the most accurate, polished controller | Gemini is strong at vector graphics again |
| SVG artwork vs Opus 5.5 | The creator rated Neon’s images more detailed and creative | Good for icons, illustrations and simple animations |
| 3D rainy Tokyo street (xhigh thinking) | Beautiful lighting and reflections, but took about 39 minutes | Top thinking levels are slow; use them sparingly |
| Animated short film of Keats’s poem “To Autumn” | Built in plain JavaScript from a single prompt | Shows creative range, not just coding |
| Minecraft-style game in the Gemini app | Worked after two tries and one error fix; survival mode missing | The chat app is usable, but Antigravity gives better builds |
The creator also used Antigravity’s /swarm feature, which splits a big task across several AI agents working in parallel. Google recommends Antigravity as the best place to use Neon for multi-step work.
On computer use (letting AI control a screen), Neon looked competitive with GPT-6.2 Sol on automation benchmarks. The creator was careful to say those tests don’t directly measure clicking around a real desktop, so real-world testing is still needed.
Try it yourself: pick one real task you did with AI this week, such as a landing page section, a product graphic or a short report, and run it through Neon and your usual model. Compare the result, the time taken and the cost. Our guide How to Try Gemini 4 Neon in Google AI Studio and Pick a Thinking Level shows you how to get started.
Where does Gemini 4 Neon fall short?
Neon is not the smartest model available, and the creator was clear about that. In his view, GPT-6.2 Astra, Claude Fable 5.2, Opus 5.5 and GPT-6.2 Sol all still have the edge on the hardest reasoning and coding tasks, although they cost more and take longer.
- Polish: on the battle scene and the shooter game, premium models produced slightly better final quality.
- Speed at the top setting: xhigh thinking can be slow, as the 39-minute Tokyo scene showed.
- One-shot results need work: several demos were single prompts. Good first drafts, but you will still refine them.
- Real-world evidence is thin: Neon is days old. The creator himself said more testing is needed.
A sensible way to use that gap is to let Neon do the bulk of the work and bring in a premium model only for the final polish. We explain how in How to Split a Project Between a Fast, Cheap AI Model and a Premium One.
Is Neon really a Flash-class model?
🔎 Unconfirmed / rumour: Before launch, Neon’s description reportedly said it was built on Gemini 4 Hydrogen using reinforcement learning, with Flash-tier speed and cost. The creator, crediting an observer on X, noticed both references were removed before release. That has led some people to ask whether Neon is really a Flash-class model priced as a mid-tier one. Google has not commented, and nobody outside Google knows its intentions.
For your decision, it matters less than you might think. What counts is the price you pay, the speed you get and whether the output is good enough for your task. Judge Neon on those, not on its label.
Should you use Gemini 4 Neon for your work?
Yes, it is worth testing if you pay for AI by usage or need quick, good-quality drafts, but don’t drop your current model on the strength of demos alone. Here is a simple way to decide.
- List your three most common AI tasks, for example blog drafts, landing page code and product images.
- Run each through Neon at the default high level and through your current model, using the same prompt.
- Score each result on quality, time taken and cost. Our guide How to Test Whether a Cheaper AI Model Is Good Enough for Your Work has a simple scoring sheet.
- Check the facts in anything factual. Lower fabrication is not zero fabrication.
- Switch task by task, not all at once. Keep your premium model for the jobs where it clearly wins.
💡 Tip: If you are on a Google AI Pro plan already, you can try Neon in the Gemini app at no extra cost. Use the model drop-down.
Frequently asked questions
When was Gemini 4 Neon released?
Google announced it on 18 November 2026, and it began rolling out the same day.
How much does Gemini 4 Neon cost?
On the Gemini API it costs $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens, including thinking tokens.
Can I use Gemini 4 Neon for free?
Not on the free Gemini API tier. You need the paid API tier, or a Google AI Pro or Ultra subscription to use it in the Gemini app, AI Studio or Antigravity.
Is Gemini 4 Neon better than Claude Opus 5.5?
Not across the board. It matched Opus 5.5 on some coding benchmarks at a lower cost and beat it on SVG graphics in the creator’s tests, but premium models still produced slightly better results on some builds.
What are Gemini 4 Neon’s thinking levels?
Low, medium, high (the default) and xhigh. Higher levels think longer, cost more and take more time.
Was the WorldofAI video sponsored?
Yes, by Mammouth, an AI platform that bundles several models from $10 a month.
The bottom line
Gemini 4 Neon is a strong value model rather than a new champion. It is fast, reasonably priced and genuinely good at graphics, 3D and front-end work, and it seems to make things up less than older Gemini models. If you pay for AI by usage, it deserves a place in your testing this week. Run it on your real tasks, keep score, and only switch where it earns it.
Related posts
- Gemini 4 Hydrogen: Google’s Fast New Flash Model, Anthropic’s “Slow Down” Pledge and This Week’s AI News
- Low, High or Max: How to Choose an AI Reasoning Effort Level
- Do You Need the Most Expensive AI Model? Why Cheaper Is Often Enough
