Requesty
Back|NOV '26AI MODELS / MODEL COMPARISONS
7 MIN READ|

Gemini Neon pricing, benchmarks and where it fits in Gemini 4

Last updated

In November 2026, Gemini 4 Neon costs $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens on the paid Gemini API. Google introduced Neon on November 18, with pricing between the introductory and standard rates for Carbon and Argon.

The date decides which model is cheaper. Neon costs 50% more per token than Carbon through November 30; when Carbon's introductory rates double on December 1, Neon becomes 25% cheaper per token.

Gemini Neon API pricing

Google's launch announcement gives Neon the standard rates below. Cached input carries a 95% discount against regular input.

Gemini 4 Neon token typePrice per million tokens
Input$3.00
Cached input$0.15
Output$15.00

Neon is available on the paid Gemini API tier. Google lists one rate for the whole 1,048,576-token context window, with no higher tier for long prompts. Thinking tokens are billed as output, so the output charge covers both reasoning tokens and the answer.

Implicit caching applies automatically to eligible shared prompt prefixes of at least 4,096 tokens. Cache hits are billed at $0.15 per million tokens. Google's model guide says Neon does not support explicit context caching.

How Neon compares on price

These prices are per million tokens. Carbon and Argon keep their introductory rates through November 30, 2026, then move to higher rates on December 1.

Model and price dateInputCached inputOutput
Gemini 4 Neon, November 2026$3.00$0.15$15.00
Gemini 4 Carbon, through November 30, 2026$2.00$0.10$10.00
Gemini 4 Carbon, from December 1, 2026$4.00$0.20$20.00
Gemini 4 Argon, through November 30, 2026$2.00$0.10$10.00
Gemini 4 Argon, from December 1, 2026$4.00$0.20$20.00
Gemini 4 Hydrogen, November 2026$1.50$0.075$7.50
Gemini 3.8 Flash, through December 2026$0.75$0.075$3.75
Gemini 3.8 Flash, from January 2027$1.50$0.15$7.50

Sources: Google Gemini API pricing and Google Cloud Model Garden pricing.

Through November 30, each Neon token type in the table costs 1.5 times the Carbon or Argon rate. From December 1, the same Neon token costs 25% less.

Hydrogen has a lower rate than Neon throughout the comparison. Flash is cheaper on regular input and output until January 2027, when those rates rise to match Hydrogen. Through December 2026, Flash and Hydrogen both charge $0.075 per million cached input tokens; from January 2027, Hydrogen's $0.075 undercuts Flash's $0.15.

The cached rate is only part of the bill: the share of a prompt that gets a cache hit changes the result. For a wider rate-card comparison, see our 2026 LLM API price guide and our look at how model list prices change.

What a task costs

Take a task that uses 10,000 uncached input tokens and 2,000 output tokens, including any thinking tokens. Each model uses the same number of tokens and completes the task successfully.

Model and price dateInput costOutput costCost per task
Neon$0.030$0.030$0.060
Carbon or Argon, through November 30, 2026$0.020$0.020$0.040
Carbon or Argon, from December 1, 2026$0.040$0.040$0.080
Hydrogen$0.015$0.015$0.030
Flash, through December 2026$0.0075$0.0075$0.015
Flash, from January 2027$0.015$0.015$0.030

For Neon, the calculation is 10,000 / 1,000,000 x $3 for input, plus 2,000 / 1,000,000 x $15 for output. That comes to $0.06 per task, or $600 for 10,000 tasks at the same token counts.

If 8,000 of those input tokens are billed at Neon's cached rate. Input cost falls from $0.03 to $0.0072, bringing the task total to $0.0372 including output. That saving depends on a cache hit for the shared prefix.

This example gives every model the same token counts. In practice, models can generate different amounts of output or need different numbers of attempts to finish a task, so measure those differences on your own workload before choosing from the rate card.

Gemini Neon benchmarks

Google's launch announcement reports these results for Neon. The scores describe the tested setups; your application can have a different cost and completion rate.

BenchmarkNeon result reported by Google
Artificial Analysis Coding Agent Index58.6, ranked first overall
Terminal-Bench 4.165.2%
FrontierCode 2.046.2%
Vals Index70.9%, ranked second

The test setup can move a coding result. On SWE-sweep, Google reports 12.8% for Neon with the mini-SWE harness and 39.0% in Google Antigravity.

Thinking level changes the result

Artificial Analysis tested Neon at four thinking levels. Higher settings scored better on its Intelligence Index and took longer per task.

Neon thinking levelIntelligence Index scoreTime per task
Low443.7 minutes
Medium558.3 minutes
High6014.8 minutes
Xhigh6121.5 minutes

High is Neon's default setting, according to Google's model guide. In the Artificial Analysis test, moving from high to xhigh adds one index point and 6.7 minutes per task.

At the high setting, Artificial Analysis also tested Argon:

Model and thinking levelIntelligence Index scoreTime per task
Neon, high6014.8 minutes
Argon, high5029.7 minutes

At high, Neon scored 10 points more than Argon and finished in about half the time.

Google says Neon at its default high setting has a 32% lower cost per task than Carbon. In Artificial Analysis' same-setting comparison, Neon high and Carbon high both score 60 on the Intelligence Index, with Neon at $3.74 per task and Carbon's high setting at $7.76 per task using standard pricing. The same post puts Neon at $6.44 per task at xhigh, about 17% less than Carbon at high.

How to call the Gemini Neon API

The Gemini API model ID is gemini-4-neon. Google's OpenAI compatibility guide shows the OpenAI SDK setup: use a Gemini API key, set the OpenAI client's base URL to https://generativelanguage.googleapis.com/v1beta/openai/, and set the model to gemini-4-neon.

For workloads that span multiple providers, Requesty has one OpenAI-compatible API in front of 600+ models, including Gemini 4 Neon. On Requesty, the model ID carries a provider prefix: google/gemini-4-neon. Our guide to calling multiple LLMs through one API covers the integration.

Which Gemini model should you try?

Verdict

Try Neon for coding and reasoning tasks where its benchmark results justify a paid model, and test the high and medium thinking settings against your own time and quality targets. Try Hydrogen when the lower token rate or free API tier is the starting point. Compare Carbon and Argon with Neon using the December 1 rates if the workload will run beyond November. For an existing Flash workload, recalculate the budget using its January 2027 rates before deciding whether to move.

Run a sample of your own tasks at the settings you expect to use. Record input, cached input, output and successful completions, then price the sample at both the November and upcoming rates.

Frequently asked questions
How much does Gemini 4 Neon cost per million tokens?
In November 2026, Gemini 4 Neon costs $3 per million input tokens, $0.15 per million cached input tokens and $15 per million output tokens on the paid Gemini API.
Is Gemini 4 Neon available on the free API tier?
No. Gemini 4 Neon is available on the paid Gemini API tier. Gemini 4 Hydrogen has a free API tier.
Is Gemini 4 Neon cheaper than Carbon?
Through November 30, 2026, Gemini 4 Neon costs 50% more per token than Carbon. From December 1, when Carbon's introductory rates double, Neon costs 25% less per token.
How does Gemini 4 Neon score on coding benchmarks?
Google reports that Gemini 4 Neon scores 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at its default high thinking setting.
How does Gemini 4 Neon compare with Argon?
At the high thinking setting, Artificial Analysis reports an Intelligence Index score of 60 for Gemini 4 Neon and 50 for Argon. Neon took 14.8 minutes per task, compared with 29.7 minutes for Argon.
What is Gemini 4 Neon's context window?
Gemini 4 Neon has a unified limit of 1,048,576 tokens across input and output. A single foreground segment can generate up to 65,536 output tokens.
What is the Gemini 4 Neon API model ID?
The Gemini API model ID for Gemini 4 Neon is gemini-4-neon. The model is available on the paid API tier.
Related reading

Start building with Requesty

One line of code. 600+ models. Full control.

Speak to founders