Models

Gemini 4 Neon: Compare Its Cost Before Switching

Gemini 4 Neon costs $3 per million input tokens. Compare its November and December bills with Carbon and Hydrogen before migrating.

Three translucent glass processing modules in mint, lavender and blue, representing a comparison of AI model tiers.

Google introduced Gemini 4 Neon on November 18, 2026, at $3 per million input tokens and $15 per million output tokens on the paid Gemini API Standard tier. That makes it more expensive per token than Carbon’s introductory price through November 30, but cheaper than Carbon’s listed price from December 1. Developers choosing among the three Gemini 4 models should compare their own completed tasks and billed tokens before switching, rather than treating Google’s cost-per-task claims as a guaranteed saving. Google’s announcement and API pricing give the figures.

What changed on November 18

Neon is a new model with the stable API ID gemini-4-neon. Google says it is available on the Gemini API’s paid tier through both generateContent and the Interactions API. Its release guide lists four thinking levels: low, medium, high (the default), and xhigh. Google positions Neon between Hydrogen and Carbon for frequent complex work. That is Google’s product positioning, not a result from a test by DailyTech.

Google also lists access through Google AI Studio for Pro and Ultra plans. The announcement describes a wider rollout to its other products, but an API price should not be mistaken for a subscription price or a promise that every account has access immediately. If you are choosing an API model, check availability in your own project before planning a migration. Our AI model tracker records the release; Google’s documentation is the source for the detailed limits and current prices.

Compare Neon, Carbon and Hydrogen at the same token count

Google’s pricing page publishes these paid Standard-tier US dollar rates per million tokens. Output includes billed thinking tokens. The Carbon change below is a published future price, not a charge already observed.

Model and period Uncached input Cached input Output, including thinking
Neon $3.00 $0.15 $15.00
Carbon through November 30, 2026 $2.00 $0.10 $10.00
Carbon starting December 1, 2026 $4.00 $0.20 $20.00
Hydrogen $1.50 $0.075 $7.50

At the same uncached input and billed output counts, Neon is 50% above Carbon’s November Standard rate, 25% below Carbon’s December rate, and twice Hydrogen’s rate. Those relationships hold for the listed input and output prices because the rates move together. They do not establish which model finishes your task more cheaply: a model may use a different number of tokens, tools or retries, and its answers may differ. Our Hydrogen cost comparison gives more background on that model.

Here is an illustrative calculation, not a measured bill. Assume one workload uses 100,000 uncached input tokens and 20,000 billed output tokens, with no cached input, tools, taxes or retries. Cost equals (input tokens ÷ 1,000,000 × input rate) + (billed output tokens ÷ 1,000,000 × output rate). Neon costs 0.1 × $3 + 0.02 × $15 = $0.60. Carbon costs $0.40 through November 30 and $0.80 from December 1. Hydrogen costs $0.30. This is a controlled price comparison, not a forecast of token use or model quality. Use the AI API cost calculator with your own counts.

Google says Neon can have a lower cost per task than Carbon in specified third-party benchmarks. That claim can coexist with Neon’s higher November token rates if successful tasks use fewer tokens or fewer attempts. The Carbon pricing deadline explainer covers that model’s scheduled price change in more detail. No independent Neon workload or invoice was measured for this article.

Check caching, thinking and output limits

Neon’s model specification gives a 1,048,576-token unified context window shared by input and output. Its default synchronous output maximum is 65,536 tokens; the document describes longer decoding up to the unified limit through a separate long-decode path. A million tokens is therefore a shared budget, not a million input tokens plus a million output tokens. The model accepts text, image, video, audio and PDF inputs, and produces text output. It does not support image generation, audio generation or the Live API.

The model page lists implicit caching only. Google’s pricing page charges $0.15 per million cached Neon input tokens, but that rate applies only to input actually reported as cached. Do not put all repeated prompts into the cached column by assumption. For example, if a bill really reports 80,000 cached tokens and 20,000 uncached tokens out of the 100,000 input tokens above, while billed output stays at 20,000, the Neon token calculation becomes 0.02 × $3 + 0.08 × $0.15 + 0.02 × $15 = $0.372. This is another illustration; it does not predict cache hits. Our prompt caching guide explains the general billing distinction.

Thinking level can also change the bill: Google’s pricing notes that thinking tokens are included in billed output. A fair trial records actual usage at the same thinking setting and counts retries. Google’s release guide says high is the default, so a silent change in thinking level can invalidate an apparent before-and-after comparison. Neon has no API free tier on the published pricing page, whereas Hydrogen’s Standard row includes one; check the applicable tier before using a free-tier assumption.

What to test before changing production

Start with a small set of your own representative tasks. Keep the current production model as the baseline, then run the same inputs through Neon with matched settings where possible. Record whether each output completes the task correctly, the billed uncached and cached input, billed output including thinking, latency, and retries. Calculate total cost ÷ successful tasks for each model and compare failure cases, rather than selecting the lowest price per token alone. This is a proposed evaluation method, not a test we performed.

If your current workload uses Carbon, repeat the calculation with the listed December rate so the decision reflects the upcoming deadline. If Hydrogen already meets your quality threshold, Neon’s higher token rate needs a measurable benefit in your workflow. Google’s Neon migration guide is the place to check API-specific steps. Nothing in the release requires an immediate move from either existing model, and this article makes no claim that one is being retired.

Sources

All pages read on November 19, 2026.

Free toolWhat will 10,000 chats a month cost?Estimate my bill
DailyTech Editorial Team

We explain AI with numbers you can check. Every price, limit and benchmark links to its source and the date we read it. How we work · Report an error