Gemini API pricing: what each model costs, the free tier and the 2027 increase

Gemini API pricing has two things the other big providers do not: a free tier that covers the Flash models, and price increases already announced for Gemini 4 Argon and Gemini 3.8 Flash. On October 28, 2026, Google's price list for the Gemini Developer API runs from $0.25 per million input tokens and $1.50 per million output tokens on Gemini 3.1 Flash-Lite to $2 and $12 on Gemini 3.1 Pro. Gemini 4 Argon, which Google released on October 13 as its most capable model, costs $2 and $10 until November 30, 2026 and $4 and $20 from December 1, and Gemini 3.8 Flash costs $0.75 and $3.75 until December 31, 2026 and $1.50 and $7.50 from January 1, 2027. This guide lists every current price, explains what the free tier really gives you, how the paid tiers and limits work and what a real workload costs, before and after the increases.

Gemini API prices per model

Table: Gemini Developer API prices per million tokens, paid tier, prompts up to 200,000 tokens (USD, checked October 28, 2026)

Five rules sit behind the table:

• Thinking is part of the output price. Google's table labels the output price "including thinking tokens", so the reasoning a model does before it answers is billed at the output rate.
• Long prompts cost more on Gemini 3.1 Pro. Above 200,000 prompt tokens it charges $4 input, $18 output and $0.40 for cached input.
• Audio input costs double on Flash-Lite: $0.50 per million tokens instead of $0.25.
• Gemini 3.6 Flash is listed at the same prices as 3.8 Flash, including the January increase, and 3.5 Flash-Lite at $0.30 and $2.50. Google deprecated Gemini 3.7 Flash on October 8, 2026 and replaced it with 3.8 Flash.
• Gemini 4 Argon only caches implicitly. Google lists no explicit cache and no storage price for it, and it is not available on the free tier.

What the free tier gives you, and what it costs you

The free tier is real money saved: Google lists input and output on Gemini 3.8 Flash and 3.1 Flash-Lite as "free of charge" on the free tier. There are three catches:

1. Your content is used to improve Google's products. The price table says it plainly: on the free plan, "Content used to improve our products"; on the paid plan, content is not used. Do not send customer data, personal data or anything confidential through the free tier.
2. The limits are not published in the documentation. Google's rate limits page says limits vary by model and sends you to the Gemini API console to see them for your project. Plan for low volumes and expect errors when you exceed requests per minute, tokens per minute or requests per day. Daily quotas reset at midnight Pacific time.
3. Gemini 4 Argon, Gemini 3.1 Pro and Google Search grounding are not in the free tier. For those you need a billing account.

The free tier is a good place to prototype and to measure how many tokens your prompts use. For a product with real users, the paid tier is the one to budget for.

Paid tiers, Batch, Flex and Priority

Processing tiers. Standard is the reference price. Batch and Flex are listed at half of it; Batch is the asynchronous option for work that can wait. Priority costs 1.8 times the standard price ($3.60 input and $21.60 output on Gemini 3.1 Pro). One detail: on Gemini 3.1 Pro the cached-input price stays at the standard $0.20 in Batch.

Usage tiers. Your rate limits rise with what you have spent, per Google Cloud billing account:

• Tier 1: link an active billing account. Billing cap of $250.
• Tier 2: $100 of cumulative paid spend and 3 days since the first payment. Cap of $2,000.
• Tier 3: $1,000 of spend and 30 days since the first payment. Cap from $20,000 to over $100,000.

Google also applies a spending limit over a rolling 10 minutes: $10 on Tier 1, $50 on Tier 2 and $200 on Tier 3. Limits apply per project, not per API key, and a request that exceeds any one of them returns an error.

Grounding with Google Search. Paid projects get 5,000 search requests a month free, shared across Gemini 3 and newer models, and then pay $14 per 1,000 search queries. Google notes that you are charged for each individual search query the model performs, so one prompt can trigger several. Grounding with Google Maps has the same free allowance and price.

What a real workload costs on each model

Take a support assistant that answers 100,000 questions a month. Each request sends 2,000 tokens of fixed instructions and reference text that can be cached, plus 1,000 new tokens, and gets back 400 tokens including thinking.

On Gemini 3.8 Flash at the current price, the month costs:

• New input: 100 million tokens × $0.75 per million = $75
• Cached input: 200 million tokens × $0.075 per million = $15
• Output: 40 million tokens × $3.75 per million = $150
• Cache storage: 2,000 tokens kept for about 730 hours × $0.50 per million per hour = about $0.73
• Total: about $241 a month, against $375 with no caching.

Table: monthly cost of the sample workload (our calculation from Google's prices, October 28, 2026)

The caching columns include storage for one cached prefix all month, except on Gemini 4 Argon, which only caches implicitly: its figures assume every request hits that cache and pay no storage. The figures leave out grounding, Priority and tax, and the Batch column only fits work that can wait. The January 1 increase doubles the bill on 3.8 Flash for the same workload, so check whether Flash-Lite passes your tests before then; the December 1 increase does the same on Gemini 4 Argon. The same token counts on Claude's API or OpenAI's are not the same text: each provider's tokenizer splits text differently, as explained in what a token is. Compare cost per task, measured on your own prompts.

How to keep a Gemini API bill under control

1. Prototype on the free tier, ship on the paid one. Use free requests to measure tokens and quality, then move real user traffic to a billing account so your data is not used to improve Google's products.
2. Test Flash-Lite before January 1. In our example it costs $135 a month without caching, against $375 on 3.8 Flash today and $750 after the increase.
3. Cache the fixed part of the prompt. Storage is cheap for a short prefix ($0.73 a month in our example), and reads cost 10% of the input price. Our prompt caching guide shows how to order a prompt for hits.
4. Keep prompts under 200,000 tokens on 3.1 Pro, where the long-prompt tier doubles the input price.
5. Watch grounding. Every search query the model runs counts, so cap how often it searches for questions that do not need fresh information.
6. Check Gemini 4 Argon before December 1. In our example it costs $620 a month with caching, less than the $727 of 3.1 Pro, and $1,240 once its price doubles.

If you mean the Gemini app and Google's Gemini plans rather than the API, those are a separate product with their own subscriptions; see ChatGPT vs Gemini for what the consumer plans include, and our LLM API price comparison, last checked on October 26, for the four main API providers side by side.