How much does the Gemini API cost? In October 2026, per million tokens: Gemini 4 Argon is $2 input and $10 output, Gemini 3.1 Pro Preview is $2 and $12 for prompts up to 200K tokens, Gemini 3.8 Flash is $0.75 and $3.75, and Gemini 3.5 Flash-Lite is $0.30 and $2.501. Unlike most rate cards, Google's already shows the next prices: Argon doubles on 1 December 2026, and 3.8 Flash and 3.6 Flash double on 1 January 2027.
If you are planning a 2027 budget on Gemini, those dates matter more than any other line on the page.
Current text model rates
| Model | Free tier | Input | Output (incl. thinking) | Scheduled change |
|---|---|---|---|---|
| Gemini 4 Argon | No | $2.00 | $10.00 | $4.00 / $20.00 from 1 Dec 2026 |
| Gemini 3.1 Pro Preview (≤200K) | No | $2.00 | $12.00 | — |
| Gemini 3.1 Pro Preview (>200K) | No | $4.00 | $18.00 | — |
| Gemini 3.8 Flash | Yes | $0.75 | $3.75 | $1.50 / $7.50 from 1 Jan 2027 |
| Gemini 3.6 Flash | Yes | $0.75 | $3.75 | $1.50 / $7.50 from 1 Jan 2027 |
| Gemini 3.5 Flash-Lite | Yes | $0.30 | $2.50 | — |
| Gemini 3.1 Flash-Lite | Yes | $0.25 | $1.50 | — |
| Gemini 2.5 Pro (≤200K) | Yes | $1.25 | $10.00 | — |
| Gemini 2.5 Flash | Yes | $0.30 | $2.50 | — |
| Gemini 2.5 Flash-Lite | Yes | $0.10 | $0.40 | — |
Google describes Gemini 4 Argon as its flagship reasoning model for complex software engineering and long multi-step agents, and it is not available on the free tier1. Gemini 3.1 Pro is still labelled Preview, with the model id gemini-3.1-pro-preview. For where Flash and Flash-Lite sit against GPT-6 Luna and Claude Haiku, see fast, low-cost models compared.
Plan for the price changes now
The changes are not promotions that might be extended; they are dated prices on the same page1:
- Gemini 4 Argon goes from $2 and $10 to $4 and $20 on 1 December 2026. Its implicit cache rate goes from $0.10 to $0.20.
- Gemini 3.8 Flash and 3.6 Flash go from $0.75 and $3.75 to $1.50 and $7.50 on 1 January 2027, and their cache, storage, Batch and Priority rates double too.
A workload that is cost-justified at today's Flash rate should be checked against the 2027 rate before it ships. If it only works at the introductory price, decide now whether the fallback is a smaller model, Batch, or a different provider, and put that decision in your 2027 budget.
The modifiers
| Modifier | Effect | Notes |
|---|---|---|
| Batch | Half the standard rate | Gemini 3.8 Flash Batch is $0.375 and $1.875 until 31 Dec 2026 |
| Flex | Half the standard rate on the models that list it | Gemini 3.8 Flash Flex matches its Batch rate |
| Priority | Above the standard rate | Gemini 3.8 Flash Priority is $1.35 and $6.75 until 31 Dec 2026 |
| Long context | Higher rates above 200K tokens | Only on Gemini 3.1 Pro and 2.5 Pro |
Context caching has two charges. Reading cached tokens is billed per million tokens (Gemini 3.8 Flash: $0.075), and explicit caches also carry a storage charge per million tokens per hour ($0.50 on 3.8 Flash until the end of 2026, then $1.00)1. A large cache kept warm all night for a handful of morning requests can cost more in storage than it saves in reads. Gemini 4 Argon lists implicit caching only.
Grounding with Google Search on Gemini 3 and later is free for 5,000 a month, shared across those models, then "$14 per 1,000 requests", and Google adds in a footnote that each individual search query performed is charged1. Gemini 2.5 models are priced differently: a daily free allowance, then $35 per 1,000 grounded prompts.
The free tier: what it costs that is not money
Google's pricing page says it plainly: on the free tier, content is used to improve Google's products; on the paid tier it is not1. That makes the free tier fine for prototypes with synthetic data and wrong for anything carrying customer data, contracts or source code you would not publish.
Rate limits also differ by tier, and they are applied per project, not per API key2. Tier 1 requires a linked billing account; Tier 2 requires $100 paid and three days since the first payment; Tier 3 requires $1,000 and 30 days. Each paid tier also has a spend rate limit per ten minutes ($10, $50 and $200)2. Creating more keys in the same project does not raise any limit; separating workloads into projects does, and it also separates their usage reports.
Counting Gemini tokens
Google's guide gives the rules of thumb: about 4 characters per token for text, 258 tokens for an image up to 384 pixels on each side, larger images split into 768 by 768 tiles of 258 tokens each, and 32 tokens per second of audio. System instructions and tool definitions count as input3. For an exact figure before you send, call count_tokens (Python) or countTokens (JavaScript) with the same contents. Count tokens before you call the API compares this with OpenAI's and Anthropic's counting endpoints.
Three requests, worked through
Choosing a Gemini model by workload
The rate card is easier to use once each model has a job:
- High-volume extraction, tagging and routing. Gemini 3.1 Flash-Lite or 2.5 Flash-Lite. Both have free tiers for development, low output rates, and no scheduled increase on the pricing page. Test them first; a task that passes on Flash-Lite never needed Flash.
- Interactive chat and agents with tools. Gemini 3.8 Flash. Its current rate is attractive, but model the workload at $1.50 and $7.50 before you commit a product margin to it, and keep a second candidate evaluated.
- Long documents and hard reasoning. Gemini 3.1 Pro Preview below 200K tokens, or Gemini 4 Argon for multi-step engineering work. Check the Preview label against your own change-management rules: a preview model can change or be withdrawn on shorter notice than a stable one.
- Anything nobody waits for. The same models on Batch, at half the standard rate.
Whichever you pick, measure cost per successful task rather than cost per token. A model that costs twice as much per token but finishes the job in one attempt instead of three is the less expensive model. Cost per successful task sets out the measurement.
Pricing Gemini requests with INFRO
You connect your own Google AI Studio or Vertex AI account, and Google bills you directly; INFRO takes no share of it. INFRO estimates each Gemini request at Google's list rate or your negotiated rate, records thinking tokens as output, and can refuse a request before Google is called if a budget for the organization, project, member or key has no room. If you hold a negotiated Gemini rate, enter it once in the console (Team plan and above) and it prices every matching request from then on, whichever service sent it.
Summary
Read Gemini's page for two things beyond the rates: the dates and the terms. Price your 2027 workloads at the 2027 rates, keep customer data off the free tier, budget explicit-cache storage by the hour, and remember that rate limits follow projects, not keys.