Gemini pricing for app plans and API costs 2026

Budgeting for Gemini? Separate app subscriptions from API token charges, then work through the costs of a catalogue-connected store assistant.

Gemini pricing starts with free personal access, while Google AI Plus costs AUD $7.99 a month and Google AI Pro costs AUD $32.99. A customer-facing store assistant has a separate API bill, calculated from the model and tokens it uses.

A personal plan can help you draft product descriptions or analyse a spreadsheet. Your storefront budget also needs catalogue retrieval and store integrations. Hosting keeps those services running.

Gemini pricing for the app and the API

Google sells personal Gemini access through Google AI subscriptions and charges developers separately for API use. An application programming interface (API) lets software send requests to a model and receive its answers.

The Australian Gemini subscription page lists these monthly prices. They buy personal app access. The API tables later in this guide use USD prices per million tokens, before applicable taxes.

Personal planAdvertised monthly priceGemini accessStorage included
FreeAUD $0Flash-Lite15 GB
Google AI PlusAUD $7.99Flash, with twice the free plan’s usage access400 GB
Google AI ProAUD $32.99Pro, Flash and Gemini 4 Argon, with four times free usage access5 TB
Google AI Ultra, lower allowanceAUD $149.99Five times Pro usage limits20 TB starting allowance
Google AI Ultra, higher allowanceAUD $329.99Twenty times Pro usage limitsHigher storage and creative allowances

These are advertised monthly amounts, without an annual-payment discount assumed. For Australian purchases using Google Play billing, Google’s tax policy includes 10% goods and services tax (GST) in tax-inclusive prices. The seller’s invoice supplies the tax breakdown for your purchase.

Google AI Pro is a subscription name. Gemini Pro is a model family. Older Gemini Advanced wording describes premium app access, so it doesn’t identify the API model or its rate.

App plans, included access and limits

Paid app plans increase personal usage allowances and add access to models and Google features. Their multiples describe relative access, rather than a promise of unlimited messages.

The free app suits occasional drafting and product research. Plus adds Flash access and more capacity. Pro adds higher model access alongside Gemini in Google apps and more storage. Ultra suits much heavier personal use, with separate allowance levels.

Creative credits apply to Google’s own tools. Google One’s terms restrict those credits to designated features. Don’t subtract them from a storefront API estimate.

Within the app, a retailer can work on catalogue descriptions, summarise supplied documents or explore shopper questions. A customer-facing assistant needs software that retrieves current products, handles shopper sessions and controls any store actions. Buying a personal subscription doesn’t supply that integration.

Use the pricing explanations directory when you need to compare another provider’s billing model alongside Gemini.

API model and token charges

The Gemini Developer API bills model usage, while Google AI Studio is the interface for exploring models and managing projects. Vertex AI is Google Cloud’s separate managed route, whose pricing page now uses the Gemini Enterprise Agent Platform name.

A token is a unit of text processed by a model. Input includes instructions, conversation history and retrieved products. Output includes the answer and any billable thinking tokens. Catalogue size alone doesn’t tell you how many tokens each conversation consumes.

Standard text rates

Google’s Developer API pricing lists the following USD amounts per million tokens. Cached input is a separate input rate. Explicit cache storage is additional.

Model identifierInputOutput, including thinkingCached inputCache storage per million tokens per hour
gemini-3.5-flash-liteUSD $0.30USD $2.50USD $0.03USD $1.00
gemini-3.1-flash-liteUSD $0.25USD $1.50USD $0.025USD $1.00
gemini-3-flash-previewUSD $0.50USD $3.00USD $0.05USD $1.00
gemini-3.1-pro-preview, prompts up to 200,000 tokensUSD $2.00USD $12.00USD $0.20USD $4.50
gemini-3.1-pro-preview, longer promptsUSD $4.00USD $18.00USD $0.40USD $4.50
gemini-2.5-flashUSD $0.30USD $2.50USD $0.03USD $1.00
gemini-2.5-flash-liteUSD $0.10USD $0.40USD $0.01USD $1.00
gemini-2.5-pro, prompts up to 200,000 tokensUSD $1.25USD $10.00USD $0.125USD $4.50
gemini-2.5-pro, longer promptsUSD $2.50USD $15.00USD $0.25USD $4.50

Gemini 3.8 Flash and 3.6 Flash have introductory input/output rates of USD $0.75/$3.75, rising to USD $1.50/$7.50 afterwards. Gemini 4 Argon’s introductory rates are USD $2/$10, then USD $4/$20. Budget each model against the applicable pricing window linked above.

Retrieval and extra charges

Standalone gemini-embedding-2 text embeddings cost USD $0.20 per million input tokens. Embeddings turn text into numerical representations for catalogue matching. Google’s managed File Search instead lists its own embedding charge of USD $0.15 per million tokens.

Gemini 3 and newer Search grounding includes 5,000 search queries monthly, shared across those models, then USD $14 per 1,000. One request can generate several queries. Searching your own catalogue doesn’t automatically invoke Google Search grounding.

Free usage and the Google Cloud route

The Gemini billing guide describes model-specific free access and paid billing accounts. The rate-limit guide separates request and token limits. A free token price doesn’t guarantee enough capacity for a busy storefront.

Paid billing is needed when the chosen model or feature requires it, or the free allowance cannot serve the workload. Google’s billing guide ties paid access to higher limits and prompts excluded from product improvement under its terms. Prepay accounts need a positive balance. Creating an API key identifies a project, but doesn’t buy a recurring token allowance.

Vertex AI pricing uses its own model, region and service rows. Google lists USD reference prices and directs other billing currencies to Cloud Platform SKUs. Regional processing and separately purchased services can change the total, so a Developer API estimate cannot double as a Vertex invoice.

Flash, pro and version-specific price requests

The exact model identifier determines the rate and availability. “Gemini 3 pricing” or “Gemini Pro pricing” leaves too much unspecified for a usable quote.

Google’s model catalogue lists stable Flash identifiers including gemini-3.8-flash, gemini-3.6-flash, gemini-3.5-flash-lite and gemini-3.1-flash-lite. It lists gemini-3.1-pro-preview and gemini-3-flash-preview as preview models. Preview is part of their identity, not an optional word in a budget.

The model lifecycle record explains several confusing names. gemini-3.5-flash routes to gemini-3.6-flash, while gemini-3.7-flash routes to gemini-3.8-flash. Neither alias is a separately selectable AI Studio model. Price the destination model.

Gemini 3 Pro’s old gemini-3-pro-preview is shut down. Gemini 2.0 Flash endpoints are also retired. Google’s release notes record the shutdown of Gemini 1.5 Pro and Flash. Old prices for these endpoints cannot fund a new deployment.

The 2.5 models remain available through the API for established users, with access restricted to people who previously used them. Google directs new projects to newer models. Their lower listed rates therefore don’t establish that a new store can choose them.

“Gemini Flash 3.5” and “Gemini 3.5 Flash Lite” need to resolve to the exact published identifier before you assign a price. “Gemini Omni” identifies a different model line. Don’t apply this text-conversation budget to video generation.

A store conversation budget

An illustrative store with 10,000 conversations a month spends USD $66.52 on its first month of Gemini 3.5 Flash-Lite tokens and embeddings under these assumptions. This is a calculation, with hypothetical usage rather than measured store traffic.

Assume each conversation needs three model requests. Every request has 4,000 input tokens, including retrieved catalogue context and history, and 400 billable output tokens. Each request also makes one catalogue retrieval and embeds a 20-token query.

First-month calculation

ItemAssumed usageCalculationCost before tax
Model input30,000 requests, 120 million tokens120 × USD $0.30USD $36.00
Model output12 million tokens12 × USD $2.50USD $30.00
Initial catalogue embeddings2 million text tokens2 × USD $0.20USD $0.40
Query embeddings600,000 tokens0.6 × USD $0.20USD $0.12
Catalogue retrieval30,000 calls to your own search serviceNo separate per-call vendor fee assumedUSD $0.00 in this model bill
First-month totalUncached text requestsUSD $36 + USD $30 + USD $0.40 + USD $0.12USD $66.52

The example uses standalone Embedding 2, without managed File Search, Search grounding or cache storage. Hosting your search service still costs money. Later months cost USD $66.12 before fresh catalogue embeddings and other operating expenses.

A hypothetical exchange rate of AUD $1.50 per USD makes the first-month subtotal AUD $99.78 before tax and currency fees. This assumption demonstrates conversion. It isn’t a live exchange-rate quote or a Google Australian tariff.

What changes the same budget

Change from the baselineNew model usageFirst-month token and embedding subtotal
Double input context to 8,000 tokens per request240 million input, 12 million outputUSD $102.52
Add one 4,000-input/400-output follow-up request to every conversation160 million input, 16 million output, 800,000 query-embedding tokensUSD $88.56
Repeat retrieval and generation for 10% more requests132 million input, 13.2 million output, 660,000 query-embedding tokensUSD $73.13

Each row changes only the named assumption. The initial catalogue embedding remains USD $0.40. Retries that complete another generation consume another allowance of billable tokens.

Caching can reduce repeated input charges when requests reuse eligible content. Its storage cost also runs while the cache exists. Cache hits, minimum sizes and retention determine whether it saves money, so the baseline assumes zero hits.

These totals buy processing, regardless of whether a shopper purchases. They aren’t a cost per sale or a conversion forecast.

Costs beyond the model bill

Building a Gemini store assistant leaves you responsible for the catalogue connection and the service surrounding the model. A low token subtotal can sit inside a much larger operating budget.

Catalogue updates need imports, stock refreshes and error handling. Store integrations need access controls for order lookups and any basket or checkout actions. The retailer also needs somewhere to run its search service and storefront integration.

Monitoring must show failed requests and spending. Maintenance covers provider changes, model replacements and testing after a catalogue or policy update. Staff time belongs in the budget even when an existing employee does the work.

Keep these amounts separate from token charges when comparing a custom build with a packaged service. The guide to whole-store AI chatbot costs covers that wider buying decision. Google’s shopper-facing discovery channel has a different merchant model, explained in the guide to AI shopping through Google.

Compare equivalent costs with ChatGPT and Claude

Compare personal subscriptions with personal subscriptions, then compare model APIs using the same workload. A monthly app allowance cannot replace the model input and output rows in a store budget.

For personal use, Google AI Pro’s advertised Australian monthly price is AUD $32.99. ChatGPT Plus lists USD $20 monthly, with API usage billed separately. Claude Pro lists USD $20 monthly in the US, with local pricing where supported.

At the illustrative AUD $1.50 exchange rate, USD $20 converts to AUD $30 before applicable tax and fees. That conversion doesn’t establish an Australian checkout price. Compare the actual monthly invoice totals on the same tax basis before ranking app costs. Annual Claude billing and Google’s included storage change that comparison.

Personal usage limits also differ. A nominally cheaper subscription can be poor value if its allowance doesn’t cover your work. Treat file analysis and app features separately from the ability to fund a storefront integration.

The identical API workload

The table uses the same 120 million input and 12 million output tokens, with every request below 100,000 input tokens. It assumes standard global processing, zero cache hits and no provider web grounding. Each row includes the same USD $0.52 of Google catalogue/query embeddings and assumes the same self-hosted retrieval service.

Provider and modelInput per million tokensOutput per million tokensFirst-month subtotal before tax
Gemini gemini-3.5-flash-liteUSD $0.30USD $2.50USD $66.52
Gemini gemini-3.8-flash, introductoryUSD $0.75USD $3.75USD $135.52
OpenAI gpt-6-lunaUSD $0.10USD $0.50USD $18.52
Claude claude-haiku-5-5USD $0.10USD $0.50USD $18.52
Gemini gemini-3.1-pro-previewUSD $2.00USD $12.00USD $384.52
OpenAI gpt-6.1-solUSD $2.00USD $10.00USD $360.52
Claude claude-sonnet-5-5USD $2.00USD $10.00USD $360.52

OpenAI’s API pricing and Anthropic’s Claude API price table supply their respective rates. Gemini 3.8 Flash’s later standard rate raises its subtotal to USD $270.52. Model names in this table identify billing choices, without claiming equal answer quality. Each provider tokenises text differently. Fixed billed-token counts don’t promise identical usage for the same text.

On these short uncached requests, GPT-6 Luna and Haiku 5.5 have the lowest listed token subtotal. Haiku’s higher rate for prompts above 100,000 tokens changes the calculation. Different output lengths or extra reasoning also change what you pay.

A cache-heavy workload needs another estimate. GPT-6 Luna lists USD $0.01 per million cached-input tokens, with paid cache-write options. Haiku 5.5 lists USD $0.01 cache reads and USD $0.125 five-minute writes for short prompts. Gemini 3.5 Flash-Lite lists USD $0.03 reads plus hourly cache storage. Charge cache misses and writes too.

Start with your expected requests and token counts, then add the integration and operating costs your store retains. For a broader replacement decision beyond the bill, read the guide to Gemini alternatives for retailers.

Give your store its own shopping agent

Customers ask in their own words. Your agent finds, compares and checks out from your catalogue, right on your site.

See it in actionFree trial · Simple setup