Google AI model guide
Estimate standard text cost across current Gemini models, then account for the parts a text box cannot measure: images, audio, video, PDFs, thinking, caching, tools, and the service tier used in production.
The right starting point depends on whether the job is hard, frequent, multimodal, or latency-sensitive. These are not interchangeable labels. Test the smallest plausible model with real examples and track failures before moving up the ladder.
Multimodal and agentic production work
Google positions 3.6 Flash as a fast, broadly capable model for agent workflows and multimodal understanding. It accepts text, images, video, audio, and PDFs, with text output.
1,048,576 input / 65,536 output
High-throughput, low-cost tasks
The economical option for extraction, classification, translation, and other workloads where throughput and unit cost matter more than top-end reasoning.
1,048,576 input / 65,536 output
Complex reasoning and richer applications
A Pro-tier option for difficult analysis and workflows that need stronger reasoning. Preview status should be part of your production risk and migration plan.
1,048,576 input / 65,536 output
Interactive cost estimate
Use text token counts and request volume to compare standard paid Gemini text rates. Multimodal inputs need Google's own token-counting tools.
Rates are per 1M tokens. Results are sorted from highest to lowest monthly estimate.
Context 1,048K / 65K
$0.0516 per request
Context 1,048K / 65K
$0.0387 per request
Context 1,048K / 65K
$0.0385 per request
Context 1,048K / 65K
$0.0349 per request
Context 1,048K / 65K
$0.009490 per request
Context 1,048K / 65K
$0.009490 per request
Context 1,048K / 65K
$0.006450 per request
Context 1,048K / 65K
$0.002080 per request
| Model | Rates per 1M | Per request | Per month |
|---|---|---|---|
Gemini 3.1 Pro Preview Context 1,048K / 65K | Input $2 Cache read $0.2 Output $12 | $0.0516 | $516.0000 |
Gemini 3.5 Flash Context 1,048K / 65K | Input $1.5 Cache read $0.15 Output $9 | $0.0387 | $387.0000 |
Gemini 2.5 Pro Context 1,048K / 65K | Input $1.25 Cache read $0.125 Output $10 | $0.0385 | $385.0000 |
Gemini 3.6 Flash Context 1,048K / 65K | Input $1.5 Cache read $0.15 Output $7.5 | $0.0349 | $349.5000 |
Gemini 2.5 Flash Context 1,048K / 65K | Input $0.3 Cache read $0.03 Output $2.5 | $0.009490 | $94.9000 |
Gemini 3.5 Flash-Lite Context 1,048K / 65K | Input $0.3 Cache read $0.03 Output $2.5 | $0.009490 | $94.9000 |
Gemini 3.1 Flash-Lite Context 1,048K / 65K | Input $0.25 Cache read $0.025 Output $1.5 | $0.006450 | $64.5000 |
Gemini 2.5 Flash-Lite Context 1,048K / 65K | Input $0.1 Cache read $0.01 Output $0.4 | $0.002080 | $20.8000 |
The estimate covers text input, eligible cached reads, and text output at standard rates. Images, video, audio, PDFs, cache storage, thinking, tools, Batch, Flex, and Priority pricing are outside this calculation.
A simple input-plus-output estimate is useful, but Gemini requests can expose more billing dimensions. Depending on the model and request, usage metadata can separate prompt content, generated candidates, cached content, internal thinking, and tool-related tokens. For multimodal work, Google's countTokens method can inspect the actual request before you send it for generation.
Instructions, conversation history, examples, JSON schemas, retrieved passages, and document text all contribute to the request. Count the assembled payload rather than only the user's latest message.
Images, audio, video, and PDFs do not map cleanly to a browser word count. Send the actual multimodal contents to Google's counting method when the result affects routing or hard limits.
Reasoning and tool use can add usage that is absent from a text-only forecast. Measure it from representative staging calls before fixing a product budget.
Gemini supports context caching, and newer model families may use implicit caching. Savings and storage charges depend on Google's rules, so this page estimates eligible reads but not the full cache lifecycle.
The calculator uses standard paid text rates so the comparison stays consistent. Google also documents other ways to run work. Batch suits asynchronous jobs and is cheaper than standard processing. Flex trades guaranteed availability for a lower price. Priority processing costs more for workloads that need more predictable service. Pick the tier before you turn a per-request estimate into a budget.
The baseline used by this calculator.
Lower-cost asynchronous processing.
Discounted, opportunistic capacity.
Higher-priced priority processing.
No. The interactive estimate on this page is for text tokens. Use Gemini's countTokens method with the actual multimodal request when non-text input affects a limit or budget.
Flash-Lite is designed for cost-sensitive throughput. Test it on your own extraction, classification, or transformation examples before choosing a more expensive model.
A production request may include thinking, cached content, tool use, media, or a different service tier. The calculator intentionally limits itself to standard text input, eligible cache reads, and text output.
Some Gemini models use higher rates above a documented input threshold. When a matching long-context rule exists in this site's pricing data, the calculator applies it automatically.