Gemini 3.1 Flash-Lite pricing
Standard pricing is $0.25 per million text, image or video input tokens, $0.50 for audio input, and $1.50 for output, including thinking tokens. A workload with 100 million text input tokens and 10 million output tokens costs $40 before caching, tools and taxes.
Calculate this $40 workload →Official API rates by processing mode
These are Google's paid Gemini API list prices in USD. Token columns are per million tokens; the final column is per million cached tokens per hour. They apply to gemini-3.1-flash-lite, the stable text-output model.
| Mode | Text / image / video input | Audio input | Output + thinking | Cache reads | Audio cache reads | Cache storage / hour |
|---|---|---|---|---|---|---|
| Standard | $0.25 | $0.50 | $1.50 | $0.025 | $0.050 | $1.00 |
| Batch | $0.125 | $0.25 | $0.75 | $0.0125 | $0.025 | $0.50 |
| Flex | $0.125 | $0.25 | $0.75 | $0.0125 | $0.025 | $0.50 |
| Priority | $0.45 | $0.90 | $2.70 | $0.045 | $0.090 | $1.80 |
The cache-read column covers text, image and video. Storage is a separate charge. The published Flash-Lite table does not list a higher price band above 200K input tokens, so do not import the Pro model's long-context pricing into this budget.
Check the model before estimating
The official model reference lists an input limit of 1,048,576 tokens and an output limit of 65,536. Flash-Lite accepts text, images, video, audio and PDFs, and returns text. Image generation, audio generation and the Live API are not supported by this model. Flash Image, Lite Image, TTS and Live models have separate billing entries.
Google describes Flash-Lite as a model for frequent, lightweight work, including translation and extraction. For a support-ticket workflow, count the system instructions and ticket text on every request. Add the generated classification or summary to output usage. A conversation that resends earlier turns can cost more than a single-message estimate suggests.
Use the stable model ID when checking an integration. A preview model name in an older tutorial is a different version reference. These prices cover the Gemini Developer API directly; a Vertex AI deployment or reseller may have different commercial terms. A consumer Gemini subscription should also be budgeted separately.
A $40 monthly budget, step by step
Assume 10,000 requests per month, each with 10,000 fresh text input tokens and 1,000 billable output tokens. The totals are 100M input and 10M output. Standard input costs 100 × $0.25 = $25; output costs 10 × $1.50 = $15. Together, that is $40, or $0.004 per request.
At exactly the same token counts, Batch or Flex costs $20 and Priority costs $72. These are arithmetic comparisons, not measured savings on completed tasks. No cache reads, storage, grounding, retries or tax are included. If every request needs one identical retry, the token bill doubles.
Open the preselected cost calculator to reproduce the $40 Standard result, then adjust tokens and monthly requests. The calculator uses text rates and excludes audio, mode adjustments, cache storage and paid tools. Its cache checkbox estimates read charges only.
For a separate audio example, 1M audio input tokens plus 100K text output tokens cost $0.50 + $0.15 = $0.65 on Standard. Do not use an audio file's byte size as its token count or apply the text input rate to the recording. Budget from the modality-specific token usage.
Cache savings depend on retention time
Suppose 80M of the example's input tokens are valid cache hits, with 20M fresh input and 10M output. Standard token charges become 80 × $0.025 + 20 × $0.25 + 10 × $1.50 = $22. That excludes storage and any initial processing needed to establish the cache.
Keeping a 100,000-token explicit cache for 24 hours adds 0.1 × 24 × $1 = $2.40 at the Standard storage rate. If the same cache stays for 30 full days, storage is $72. A low read price alone does not make a long-lived cache economical. See Google's caching guide for eligibility, token minimums and lifetime settings.
Choose the mode for the deadline
Batch processing is asynchronous, with a target turnaround of 24 hours. It suits work that can wait, such as a queued document archive. Flex and Priority are separate inference options; check their latency and availability terms before changing an interactive application. Their listed token prices do not establish an application-level speed guarantee.
Search grounding can add another bill. Google's table lists 5,000 free search requests per month shared across Gemini 3 and newer models, then $14 per 1,000. Its footnote says each individual search query is charged, and a model request can generate several queries. Track those separately from token spend; the free allowance is not a new allocation for each model.
Free access and a production spending limit
The Standard pricing table lists a free tier for input and output, but not context caching. Availability and rate limits still apply. In an earlier Reddit discussion about the preview, a developer asked where to find the free and paid limits. That historical experience is not a quota promise for the current stable model: check the active project's limits in AI Studio.
Before production use, review Google's billing documentation and set a project spend cap where available. Google labels project caps experimental: billing can lag by around ten minutes, allowing overages, and invoiced accounts are excluded. Keep budget alerts as a separate notification layer. Google's price table distinguishes free-tier data use for product improvement from the paid tier; review the applicable terms before sending sensitive material.
For a budget review, record the model ID, processing mode, input modality, fresh and cached token totals, billable output and cache lifetime. Compare that estimate with actual usage, including unsuccessful attempts that incur charges. The Google API pricing catalog and token calculator provide the next comparison when the workload changes.