Artificial Intelligence (AI)

Gemini 4 Neon: Google’s New Middle Model, What It Costs and When to Use It

A night photograph of a glowing cyan and violet tube light sculpture bent into an abstract loop on an old red brick wall, its light reflected on the wet cobblestones below, a visual for Gemini 4 Neon, the new middle model in Google's Gemini 4 family, priced between Hydrogen and Carbon in the Gemini API

Gemini 4 Neon is Google’s new middle model in the Gemini 4 family, released as generally available in the Gemini API on November 18, 2026. It costs $3 per million input tokens and $15 per million output tokens, on the paid tier only. Google says it sits between Gemini 4 Hydrogen and Gemini 4 Carbon.

This is the fourth Gemini 4 model since October. We covered Gemini 4 Argon’s general availability, then Gemini 4 Hydrogen and Gemini 4 Carbon this week. This post explains where Neon fits, how the December 1 price change makes it the middle option, and what to check before you move a site chatbot or plugin. Facts are as of November 20, 2026.

The short version: Gemini 4 Neon costs twice as much as Hydrogen per token. It costs 50% more than Argon and Carbon do today. But Argon and Carbon double to $4 and $20 on December 1, 2026. From that date, Neon is 25% cheaper per token than either. If you run Argon or Carbon on work that does not need the top model, test Neon this week.

What Google released with Gemini 4 Neon

The Gemini API release notes list gemini-4-neon as generally available on November 18. The entry names the $3 and $15 prices, four thinking levels and a unified 1M-token context window.

Google’s launch post by Koray Kavukcuoglu describes Neon as "bringing frontier-level intelligence with fast end-to-end task completion." It says Carbon handles the hardest long projects, Hydrogen handles fast everyday work, and "Neon balances depth and efficiency."

According to the launch post, Neon is rolling out in these places:

  • Developers: the Gemini API paid tier in Google AI Studio and Android Studio.
  • Subscribers: Google AI Pro and Ultra subscribers can pick it in the Gemini app and try it in AI Studio. Paid Pro and Ultra subscribers can use it in Google Antigravity.
  • Businesses: Gemini for business, through the API or the model drop-down, and Antigravity.
  • Security teams: a cyber version for eligible members of Google’s Fairwind Program, over the next few days.

Google’s Gemini 4 Neon model card adds two useful facts. Neon is adapted from Gemini 4 Carbon, and its knowledge cutoff is March 2026. The card lists text, images, audio and video as inputs.

Google’s posts do not mention a separate date for Vertex AI or the Gemini Enterprise Agent Platform. If your client runs Gemini through Google Cloud, check that console before you plan a move.

Where Gemini 4 Neon sits next to Hydrogen, Argon and Carbon

Google’s What’s new in Gemini 4 Neon guide gives a plain rule. Use Carbon for the hardest architecture, debugging, science and security problems. Keep Hydrogen as the default for everyday coding and agents. Use Neon for complex work where you want speed too.

The prices below come from the Gemini API pricing page, read November 20, 2026.

Model Paid price per 1M tokens (input / output) Free tier Google’s suggested use
Gemini 4 Hydrogen $1.50 / $7.50 Yes Everyday coding and agent work
Gemini 4 Neon $3 / $15 No Complex, high-frequency work
Gemini 4 Argon $2 / $10 through Nov 30; $4 / $20 from Dec 1 No Complex work with less token use than Carbon
Gemini 4 Carbon $2 / $10 through Nov 30; $4 / $20 from Dec 1 No Hardest debugging, science and security work

Cached input on Neon costs $0.15 per million tokens. The guide says implicit caching runs on prompts of at least 4,096 tokens. Neon does not support explicit caching. Our prompt caching explainer covers how to order a prompt so a fixed block of instructions gets the discount.

The Neon guide compares Neon only with Hydrogen, Carbon and Gemini 3.5 Flash-Lite. It does not say where Argon fits now. The pricing page still calls Argon Google’s efficient Gemini 4 frontier model, with lower token spend per turn than Carbon.

Why December 1 makes Gemini 4 Neon the middle option

The pricing page shows Argon and Carbon at $2 and $10 through November 30, 2026. From December 1, both rise to $4 input and $20 output. Neon’s $3 and $15 price has no end date on the page.

Here is a worked example. Say a nonprofit’s help bot sends 4,000 input tokens per answer and gets back 800 output tokens, thinking included. These are our assumptions, and each model will use a different number of thinking tokens on the same question.

Model and price period Cost per answer 10,000 answers a month
Gemini 4 Hydrogen $0.012 $120
Gemini 4 Neon $0.024 $240
Argon or Carbon, through Nov 30 $0.016 $160
Argon or Carbon, from Dec 1 $0.032 $320

At equal token counts, Neon saves a quarter of the bill against Argon or Carbon from December. Google also claims a gap in tokens per task. The launch post says Neon at its default high level has a "32% lower cost per task than Carbon." That is Google’s figure on its chosen benchmarks. Measure your own tokens per answer before you rely on it.

The bigger saving for small sites may be Hydrogen. It costs half of Neon per token and has a free tier. If Hydrogen answers your questions well, it stays the cheapest Gemini 4 choice.

Thinking levels and output limits

Gemini 4 Neon has the same four thinking levels as Hydrogen: low, medium, high and xhigh. High is the default. Argon and Carbon default to medium and stop at high, per Google’s guide.

The guide says to start with high for most complex tasks and save xhigh for hard, multi-step problems. It warns that xhigh "increases latency and token use." Thinking tokens bill as output at $15 per million. So for a simple question bot, test low and medium first.

Output works the same way as on Hydrogen:

  • One window: input, thinking and output share 1,048,576 tokens.
  • Per request: up to 65,536 output tokens in a normal call.
  • Longer output: set background to true and raise max_output_tokens to go past 64K, up to 1M.
  • Timeouts: the guide says hard prompts can take several minutes, and suggests client timeouts of at least 15 minutes.

Moving to Gemini 4 Neon from another model

If you run another Gemini 4 model, Google says the switch needs no request schema changes. Change the model ID to gemini-4-neon, check the thinking level and make sure the project is on the paid tier.

Moving from Gemini 3.8 Flash or older takes more work. The guide’s checklist matches the one for Hydrogen:

  1. Change the model ID to gemini-4-neon.
  2. Raise low output caps. Remove hardcoded limits such as 8,192, or raise them to 65,536.
  3. Set the thinking level. Neon rejects minimal with an error.
  4. Drop explicit caching. Remove cached_content fields and rely on implicit caching.
  5. Remove sampling parameters. Delete temperature, top_p and top_k. Out-of-range values return a 400 error.
  6. Declare tools natively. Raw JSON tool calls with no declared tools can return MALFORMED_FUNCTION_CALL.
  7. Keep thought signatures in multi-turn calls, or use previous_interaction_id in the Interactions API.

Our post on which Gemini model to use predates all three new models, so treat its picks as out of date.

What Google claims about Gemini 4 Neon’s quality

The model card compares Neon with Claude Opus 5.5 and GPT-6.2 Sol. These are Google’s reported numbers, not independent tests.

Benchmark Gemini 4 Neon Claude Opus 5.5 GPT-6.2 Sol
Artificial Analysis Index v5.0.1 61 59 57
Terminal-bench 4.1 65.2% 65.8% 61.5%
FrontierCode 2.0 46.2% 47.3% 43.8%
Chartography 73.5% 66.3% 75.1%

So Google’s own table shows Neon slightly behind Opus 5.5 on two coding tests and behind GPT-6.2 Sol on chart reading. It leads on most other rows in the card.

The launch post also says Neon had its "lowest fabrication rate yet" in internal Antigravity testing. The model card still lists hallucination as a known limit. Keep a human check on anything Neon writes for members or the public.

Similar moves at OpenAI and Anthropic

All three labs now sell a middle model below their top tier. OpenAI’s GPT-6.2 Sol model page lists $2 input and $10 output per million tokens, with a 1,050,000-token context window. We covered it in GPT-6.2 Astra and Sol. Anthropic’s pricing page lists Claude Sonnet 5.5 at $2 and $10, and Claude Opus 5.5 at $4 and $20. See our posts on Claude Sonnet 5.5 and Claude Opus 5.5. Per token, Neon costs more than Sol and Sonnet 5.5 and less than Opus 5.5.

What to do before December 1

  1. List every place you call Argon or Carbon. Chatbots, CMS plugins and scheduled jobs all count.
  2. Run Gemini 4 Neon and Hydrogen on twenty real tasks from each workload. Record tokens per task and the quality of each answer.
  3. Keep Carbon only where it wins. After December 1, every Carbon call costs a third more per token than Neon.
  4. Check the paid tier. Neon has no free tier, so a free-tier key will not reach it.
  5. Set a thinking level on purpose. Do not let the high default run on simple questions.
  6. Set a project spend cap in AI Studio before the price change lands. Google added project-level caps in March 2026, per the release notes.

Frequently Asked Questions

What is Gemini 4 Neon?

It is Google’s middle Gemini 4 model, released as generally available on November 18, 2026. Google says it sits between Hydrogen and Carbon for complex work that also needs speed.

How much does Gemini 4 Neon cost?

$3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens, thinking included. The price is the same across the full 1M-token window.

Is there a free tier for Gemini 4 Neon?

No. Google’s pricing page lists Neon as not available on the free tier. Hydrogen is the only Gemini 4 model with a free tier.

Is Neon cheaper than Gemini 4 Argon?

Not until November 30. Argon costs $2 and $10 through that date, then $4 and $20 from December 1, 2026. From then on, Neon is 25% cheaper per token.

Should I switch from Carbon to Neon?

Test first. Google says Neon at high costs 32% less per task than Carbon, and its guide still points the hardest debugging, science and security work to Carbon.

Can I use Gemini 4 Neon in the Gemini app?

Yes, if you have Google AI Pro or Ultra. Google says those subscribers can pick Neon from the model drop-down in the Gemini app.

What context window and output limits does Neon have?

One 1,048,576-token window shared by input, thinking and output. A normal request returns up to 65,536 tokens. Background execution can run longer output up to 1M tokens.

What breaks when I move from Gemini 3.8 Flash?

Explicit caching, the minimal thinking level and sampling parameters such as temperature. Low output caps can also cut answers short. Google’s migration checklist lists each change.

How does Neon compare with GPT-6.2 Sol and Claude Opus 5.5?

Google’s model card shows Neon ahead on most benchmarks it lists, slightly behind Opus 5.5 on Terminal-bench 4.1 and FrontierCode 2.0, and behind GPT-6.2 Sol on Chartography. Those are Google’s numbers.

Digital Matters

Artificial Intelligence (AI) Desk