How Gemini API pricing works
Google bills the Gemini API per token, with separate prices for input and output. Output prices include thinking tokens, so a model that reasons at length before answering costs more than its list price suggests. There are four current Gemini models on the paid API:
- Gemini 4 Argon ($2.00 / $10.00): Google's flagship reasoning model, on the paid API since 13 October 2026. The price is introductory: from 1 December it doubles to $4 / $20. Not on the free tier.
- Gemini 3.1 Pro ($2.00 / $12.00): still labelled a preview. Prompts over 200,000 tokens cost $4 / $18, and the higher rate applies to the whole request.
- Gemini 3.8 Flash ($0.75 / $3.75): Google's most capable Flash model, for agents, coding and enterprise workflows. Also an introductory price, until 31 December 2026; from 1 January 2027 it doubles to $1.50 / $7.50.
- Gemini 3.5 Flash-Lite ($0.30 / $2.50): the low-cost option for high-volume work such as translation and simple data processing.
Two of the four are introductory prices that double, Gemini 4 Argon's on 1 December 2026 and Gemini 3.8 Flash's on 1 January 2027, so budget at the later rates for anything that will run past them. For Google's open-weight Gemma models, we list the cheapest third-party hosts on the Gemma page.
The free tier
Unlike Anthropic and OpenAI, Google offers a free tier on its API. Gemini 3.8 Flash and Gemini 3.5 Flash-Lite are free to use through Google AI Studio and the API, within rate limits. Gemini 4 Argon and Gemini 3.1 Pro aren't available on the free tier.
The catch is data use: on the free tier, Google may use your prompts and responses to improve its products. On the paid tier it doesn't. That makes the free tier fine for prototypes and personal projects, and a poor choice for customer data. Rate limits change often, so check Google's rate-limit table rather than relying on a copy.
Context caching: when it pays
Gemini charges for explicit context caches in two ways: a reduced per-token rate when the cache is read, plus a storage fee for every hour the cache exists.
| Model | Cache read per 1M | Storage per 1M per hour | Reads an hour to break even |
|---|---|---|---|
| Gemini 3.1 Pro | $0.20 | $4.50 | 2.5 |
| Gemini 3.8 Flash | $0.075 | $0.50 | 0.74 |
| Gemini 3.5 Flash-Lite | $0.03 | $1.00 | 3.7 |
The last column is calculated from the rates above: the storage fee divided by what each read saves against sending the same tokens as normal input. A cache on Gemini 3.8 Flash pays off if it's read about once every 80 minutes; on Gemini 3.1 Pro it needs a read every 24 minutes, and on Flash-Lite every 16. That makes caching a clear win for a chatbot's system prompt under steady traffic and a loss for a document you query twice a day. On 1 January 2027 Gemini 3.8 Flash's cache reads double to $0.15 and its storage fee to $1.00 per 1M tokens per hour, along with its other prices, so its break-even stays the same.
Gemini 4 Argon works differently: it supports implicit caching only, so repeated prompt prefixes are billed at its cache rate ($0.10 per 1M until 30 November) automatically, with no storage fee and no way to create a cache yourself.
The Batch API and Flex processing both cost 50% less than standard, in exchange for slower results, though on Gemini 3.1 Pro cache reads cost the same in Batch as in Standard. For a chat assistant handling 100M input and 20M output tokens a month, Gemini 3.8 Flash costs $150 at standard rates and $75.00 through Batch.
Thinking tokens are billed as output
Every current Gemini model can reason before it answers, and those thinking tokens are billed at the output rate even though you don't see them in the reply. A request that thinks for 5,000 tokens and answers in 500 is billed for 5,500 output tokens: about $0.02 on Gemini 3.8 Flash and $0.06 on Gemini 4 Argon at today's prices. Gemini 4 Argon can also write up to 1,048,576 output tokens with long decode, so cap the output on requests where you don't need it.
Subscription or API?
Google AI subscriptions (AI Plus, AI Pro and AI Ultra, from $4.99 a month) cover the Gemini app and Google's own tools such as Jules and Antigravity. Gemini API usage is billed separately, through Google AI Studio or Google Cloud. For your own apps and automations, use the API; our subscription vs API calculator compares the two for personal use.
Which Gemini model, and which rival?
- Most apps and agents: Gemini 3.8 Flash, especially while the introductory price lasts. Compare it with Claude Sonnet 5.5 and GPT-6.1 Sol.
- Long reasoning, agents and very long outputs: Gemini 4 Argon, which costs the same per token as GPT-6.1 Sol until 30 November. Argon or 3.1 Pro?
- High-volume, simple work: Gemini 3.5 Flash-Lite, or its rival GPT-6 Luna.
- Prototypes: the free tier of Flash or Flash-Lite, as long as you don't send private data.
Other charges
- Grounding with Google Search: 5,000 free search requests a month, shared across Gemini 3 and newer models, then $14 per 1,000.
- Grounding with Google Maps: the same allowance and price.
- Priority inference costs more for faster, more reliable responses: 1.8× standard on Gemini 4 Argon.
- Vertex AI: Gemini is also sold through Google Cloud's Vertex AI, which has its own price list and enterprise terms; this page covers the Gemini Developer API.
Sources: Gemini API pricing, Gemini 4 Argon model page, Gemini API changelog; checked 22 October 2026. Google AI plan prices: Google AI plans, checked 13 October 2026.
