aiapicostcalcUnit economicsPricesBrandsCompareBenchmarksSkills & PluginsFind my modelGuidesChanges

Gemma API pricing: free tier in the Gemini API, paid on Vertex AI

Gemma 4 costs nothing in the Gemini API free tier, where input, output and context caching are all free of charge, and $0.15 per 1M input and $0.60 per 1M output tokens on Vertex AI, with cached input at $0.015. The calculator below prices your own token counts at the Vertex rates; the free tier is rate-limited and its data terms differ (checked 2026-11-09).

Data checked 2026-11-09 · Sources: Gemini API pricing · Vertex AI pricing · Report an error

1
Cost per call$0.0006
Per 1,000 calls$0.60
Monthly bill$6.00
Same calls on the Gemini API free tier$0.00

Formula: cost per call = (uncached input × $0.15 + cached input × $0.015 + output × $0.60) ÷ 1,000,000. The monthly bill is calls × cost per call. Rates are Vertex AI list prices for Gemma 4 26B, read from Google’s pricing page on 2026-11-09; the free tier price is $0 because Google lists it as free of charge.

What Gemma costs on each route

Google publishes Gemma in two places, and they are priced in different units. The Gemini API runs Gemma on a free tier with no paid option; Vertex AI bills the same family per token. A reseller invoice is not a third published price, so it is not in this table.

RouteModelInput /1MCached input /1MOutput /1MHow it is billedOfficial source, read 2026-11-09
Gemini API, free tierGemma 4Free of chargeFree of chargeFree of chargeRate-limited free tier; paid tier listed as “Not available”Gemini API pricing
Vertex AIGemma 4 26B$0.15$0.015$0.60Pay per token at list priceVertex AI pricing
Your own GPUs, open weightsGemma 4GPU cost—GPU costGPU hours, not tokensGemma 4 model card

“Free of charge” and “Not available” are Google’s own words in the Gemma 4 table of the Gemini API pricing page. The Vertex rates are the single Gemma row on Google’s Vertex AI pricing page. The site does not add a blended or estimated price for either route.

Gemma 4 in the Gemini API is free, with two conditions

Free of charge, and no paid tier to buy

On the Gemini API pricing page, the Gemma 4 row prices input, output, context caching and context-caching storage at “Free of charge”. The paid-tier column beside those rows reads “Not available”, so there is no per-token Gemma rate in the Gemini API to upgrade to when the free tier is not enough. Tuning and grounding with Google Search are also listed as “Not available” on that row, so they are not add-ons you can buy on this route.

What free means for data and rate limits

The same table answers the question that decides most production use: free-tier traffic is listed as “Used to improve our products: Yes”, while the paid-tier column says No. If your prompts cannot be used that way, the free route is the wrong route whatever it costs. The free tier is also rate-limited; Google publishes the current limits on its own rate-limits page, and they change, so read them instead of copying a number from a round-up post.

Gemma 4 26B on Vertex AI: $0.15 in, $0.60 out

Vertex AI is where Gemma becomes a metered API. Google gives one input rate, one output rate and one cache-hit rate per 1M tokens, and the model card explains the rest of the bill: the medium Gemma 4 models take up to 256K tokens of context and the small ones 128K, so a long document is one large input charge per pass.

TypePrice per 1M tokens (USD)What it covers
Input$0.15Prompt tokens not served from cache
Cache hit$0.015Input tokens served from the context cache
Output$0.60Everything the model writes

The cache-hit rate is one tenth of the input rate, so cached input is billed 90% cheaper than fresh input. The Gemma row lists these three rates only; it does not publish a separate long-context or batch tier, so this page does not invent one.

What 1,000 and 1,000,000 calls cost

The same arithmetic the calculator runs, at the Vertex rates, for a call that sends 2,000 input tokens and gets 500 output tokens back. Every figure is calls multiplied by the published rates.

CallsInput tokensOutput tokensCostArithmetic
1,0002,000,000500,000$0.602M × $0.15 + 0.5M × $0.60
10,00020,000,0005,000,000$6.0020M × $0.15 + 5M × $0.60
100,000200,000,00050,000,000$60.00200M × $0.15 + 50M × $0.60
1,000,0002,000,000,000500,000,000$600.002B × $0.15 + 0.5B × $0.60
10,000, half the input cached10,000,000 fresh + 10,000,000 cached5,000,000$4.6510M × $0.15 + 10M × $0.015 + 5M × $0.60

Caching half the input saves $1.35 on a 10,000-call month. The free tier changes this table to zero on every row, subject to the rate limits and the data terms above.

Gemma 4 sizes and context that move the bill

Gemma 4 is not one model. The model card lists dense models from E2B to 31B and a 26B mixture-of-experts variant with 25.2B total and 3.8B active parameters, plus 128K context on the small models and 256K on the medium ones. On Vertex AI the rate is per token, so context length is the lever: a 200,000-token document costs $0.03 of input per pass at $0.15 per 1M, and repeating that across 10,000 requests costs $300.00 before any output.

When self-hosting beats both routes

Gemma 4 is open-weights, so the paid option is not limited to Vertex AI: you can run the same model on your own GPUs, where the bill is GPU hours rather than tokens. That wins when utilization is high for long stretches and loses when a single job sits idle on a rented card. The self-host vs API break-even calculator takes your GPU price and throughput and shows the utilization where self-hosting beats a per-token rate; for Gemma, put the Vertex cache-hit rate in for cached prompt tokens and compare it with your own hardware, not with a cloud GPU list price you have not read.

Other model pricing pages

The Gemini API cost calculator prices Google’s text models per call, and the per-model price list carries every verified rate with its check date. If you already know the token count, the token cost calculator converts 1K, 100K or 1M tokens into dollars; the LLM cost glossary defines cached input, blended price and the other billing terms used here; and how prices are verified explains why a rate without an official source is left out rather than estimated.

Questions

Is the Gemma API free?

On the Gemini API, yes: Google's pricing page lists Gemma 4 input, output and context caching as Free of charge and the paid tier as Not available (read 2026-11-09). Vertex AI bills the same family per token instead, at $0.15 input and $0.60 output per 1M tokens for Gemma 4 26B, with cached input at $0.015.

How much is Gemma 4 26B per 1M tokens on Vertex AI?

$0.15 for input, $0.60 for output and $0.015 for cached input, the single Gemma row on Google's Vertex AI pricing page read on 2026-11-09. A call with 2,000 input and 500 output tokens is $0.0006, which is $0.60 per 1,000 calls and $6.00 per 10,000 calls.

Can I run the free Gemma tier in production?

Two conditions decide it, and neither is about price. The free tier is rate-limited, so sustained traffic can be throttled, and Google lists free-tier data as used to improve its products while paid-tier data is not. If either condition is unacceptable, the paid alternatives are the per-token Vertex AI rate or self-hosting the open weights.

Why is there no paid Gemma tier in the Gemini API?

Because Google lists the paid-tier column for Gemma 4 as Not available on its Gemini API pricing page read 2026-11-09. The published paid route for Gemma is Vertex AI, priced per token; this page does not assume a Gemini API paid rate that Google has not published.

How much does cached Gemma input cost?

On Vertex AI a cache hit is $0.015 per 1M tokens, one tenth of the $0.15 input rate, so cached input is 90% cheaper than fresh input. For 10,000 calls of 2,000 input and 500 output tokens, caching half the input moves the bill from $6.00 to $4.65.

How does Gemma 4 compare with self-hosting it?

Gemma 4 is open-weights, so both are real options: Vertex AI charges per token, while self-hosting charges GPU hours. The break-even depends on utilization and on your own hardware price, so this site points to the self-host vs API calculator rather than quoting a single cheaper answer.

What context window does Gemma 4 have?

The Gemma 4 model card gives 128K tokens of context to the small models and 256K to the medium models, and lists dense sizes from E2B to 31B plus a 26B mixture-of-experts variant with 25.2B total and 3.8B active parameters. Context length matters to the bill because every input token is charged at the input rate.

Next step → Check the utilization where self-hosting wins

Price data last checked 2026-11-18 against official sources · how we verify

Price-change alerts

One short email when a provider changes a price or a promo ends. No spam, unsubscribe any time.