Google refreshed its Flash tier on July 21, 2026, and Gemini 3.5 Flash-Lite is the budget end of it. It is the cheapest, fastest Gemini model you can call today, built for high-volume, latency-sensitive work such as classification, extraction, short chat replies, and simple retrieval answers where throughput and cost matter more than deep reasoning. If you are sending millions of small requests per day, this is the tier Google wants in your hot path.
The refresh shipped three models at once, and the naming is easy to misread. This guide covers the model ID, pricing, benchmarks, selection criteria, and a practical way to test the endpoint.
What is Gemini 3.5 Flash-Lite?
Gemini 3.5 Flash-Lite is the smallest and cheapest model in Google’s Gemini family. It is tuned for speed and request volume rather than difficult reasoning. Google clocks it at roughly 350 output tokens per second, so responses can feel near-instant even under load.
Use this model ID:
gemini-3.5-flash-lite
Use Flash-Lite for repetitive, high-frequency tasks:
- Tagging or routing support tickets
- Extracting fields from documents
- Returning JSON from unstructured text
- Answering short questions from retrieved context
- Powering a chat widget that needs low latency
It landed as part of a three-model Flash refresh:
- Gemini 3.6 Flash: the workhorse model
- Gemini 3.5 Flash-Lite: the low-cost, high-throughput model
- Gemini 3.5 Flash Cyber: a gated security model
Read Google’s announcements on the Google blog and the DeepMind Flash page. Flash-Lite sits at the bottom of the stack on price and at the top on raw speed.
Why is it 3.5 instead of 3.6?
The new workhorse model is Gemini 3.6 Flash, but Flash-Lite and the gated Cyber model stayed on version 3.5. One launch shipped three models with two version numbers.
This is not a typo. Google uses the version number to identify the model line, not the launch event. The workhorse received the larger generational bump and the 3.6 label, while Flash-Lite remained aligned with the 3.5 line.
Also, Gemini 3.5 Flash-Lite is not the older Gemini 3.1 Flash-Lite. They share a family name but are different models. Before changing production traffic, verify that your configuration uses:
gemini-3.5-flash-lite
Specs and pricing
All prices are per million tokens through the Gemini API.
| Attribute | Value |
|---|---|
| Model ID | gemini-3.5-flash-lite |
| Input price | $0.30 / 1M tokens |
| Output price | $2.50 / 1M tokens |
| Speed | ~350 output tokens/second |
| Free tier | Yes, through Google AI Studio; rate-limited |
| Context caching | $0.03 / 1M tokens + $1.00 / 1M/hr storage |
At $0.30 per million input tokens and $2.50 per million output tokens, Flash-Lite is the cheapest published model in the current Gemini lineup.
For comparison, Gemini 3.6 Flash costs $1.50 per million input tokens and $7.50 per million output tokens. That makes Flash-Lite roughly one-fifth of the input cost and one-third of the output cost.
Check the Gemini API pricing docs before setting production budgets, because Google updates that page directly.
Use context caching for repeated prompts
Context caching is useful when many calls share the same content, such as:
- A fixed system prompt
- Product documentation
- A policy document
- A long reference file used across multiple requests
You pay to cache tokens plus an hourly storage fee, but repeated cached input is cheaper than resending the same prompt on every request.
How well does it perform?
Gemini 3.5 Flash-Lite scores 54% on Terminal-Bench 2.1, up from 31% on its predecessor. That is a meaningful improvement for a Lite-tier model.
Its best-fit workloads are still straightforward:
- Text classification
- Structured extraction
- Short chat responses
- Simple retrieval-augmented generation (RAG)
- High-volume input routing
For example, use a constrained prompt when extracting structured fields:
Extract the following fields from the support ticket:
- customer_name
- product
- issue_type
- urgency
Return valid JSON only.
Flash-Lite is not the right choice for difficult agentic coding, long multi-step tool chains, or deep reasoning tasks. If a workflow must plan across many steps, reliably sequence tool calls, or reason through a complex codebase, use Gemini 3.6 Flash or a larger model.
A lower token price does not help if retries, incorrect routing, or bad outputs cost more than the savings. Match the model tier to the task.
Where Google uses Flash-Lite
Google is rolling Flash-Lite into Google Search. That is a useful operational signal: Search-scale usage requires low latency and a low cost per request.
For developers, the takeaway is practical. The same characteristics that fit Search—high throughput and low cost—also fit production workloads such as:
- Ticket classification pipelines
- Content moderation pre-filters
- Form and document extraction
- Lightweight customer-facing chat
- RAG answer generation over short retrieved passages
Flash-Lite vs. Gemini 3.6 Flash: which should you pick?
Choose Flash-Lite when the task is simple, high-volume, and latency-sensitive. Choose Gemini 3.6 Flash when the task needs stronger reasoning, coding, or multi-step agent behavior.
| Workload | Recommended model |
|---|---|
| Ticket tagging and routing | Gemini 3.5 Flash-Lite |
| JSON extraction | Gemini 3.5 Flash-Lite |
| Short RAG answers | Gemini 3.5 Flash-Lite |
| High-volume chat replies | Gemini 3.5 Flash-Lite |
| Agentic coding | Gemini 3.6 Flash |
| Multi-step tool workflows | Gemini 3.6 Flash |
| Complex reasoning | Gemini 3.6 Flash |
A practical implementation pattern is difficulty-based routing:
- Send the common, predictable requests to Flash-Lite.
- Validate the response against your expected schema or confidence criteria.
- Escalate ambiguous or failed requests to Gemini 3.6 Flash.
- Log escalation rates so you can tune prompts and routing thresholds.
This keeps most traffic inexpensive while reserving stronger reasoning for requests that need it. See Gemini 3.5 Flash-Lite vs. 3.6 Flash for a full comparison of speed, price, and quality.
How to access and test it
Flash-Lite runs on the Gemini API. Start in Google AI Studio:
- Create or retrieve an API key.
- Set the model to
gemini-3.5-flash-lite. - Test representative prompts.
- Measure latency, output quality, and token usage before routing production traffic.
The free tier is rate-limited and Google may use free-tier data to improve its products. Avoid sending sensitive production data through a free-tier key. For endpoint and request details, use the Gemini API docs.
Test a representative request
Use a small, deterministic test case first. For example, a classification prompt should clearly specify allowed labels:
{
"contents": [
{
"parts": [
{
"text": "Classify this support ticket as exactly one of: billing, technical, account.\n\nTicket: I was charged twice for my subscription."
}
]
}
]
}
For extraction workloads, assert both the HTTP response and the output format. Do not only check that the model returned 200 OK.
Validate the API request with Apidog
Once the request works, save it as a regression test. API pricing, rate limits, and response shapes can change as providers update their APIs, and fast models still need quality checks.
With Apidog, you can:
- Create a
POSTrequest to the Gemini endpoint. - Store the Gemini API key in an environment variable rather than in the request body.
- Add assertions for the HTTP status code.
- Assert the JSON fields your application depends on.
- Save the request as a repeatable regression test.
Useful checks include:
Status code is 200
Response contains generated content
Response JSON matches the fields required by the application
Then schedule the API test to catch broken responses or behavioral drift before users encounter them.
Apidog does not run the model or replace the Gemini API. It is the client layer for calling, validating, and monitoring the endpoint alongside the rest of your API stack.
FAQ
Is Gemini 3.5 Flash-Lite the same as Gemini 3.6 Flash?
No. They are separate models from the July 21, 2026 refresh. Flash-Lite is the cheapest and fastest tier for simple, high-volume work. Gemini 3.6 Flash is the pricier workhorse for stronger reasoning and coding.
Why is it 3.5 and not 3.6?
Google used mixed versioning for this launch. Gemini 3.6 Flash received the 3.6 label, while Flash-Lite and the gated Cyber model stayed on the 3.5 line. The version identifies the model line, not the launch date.
Is this the old Gemini 3.1 Flash-Lite?
No. Gemini 3.5 Flash-Lite is a newer model. Check for the exact model ID:
gemini-3.5-flash-lite
Is Gemini 3.5 Flash-Lite free?
Google AI Studio provides a rate-limited free tier for testing and light use. Production workloads require a paid API key, and Google may use free-tier data to improve its products.
What is Flash-Lite best at?
It is best for classification, extraction, short chat replies, and simple RAG answers at high volume. For difficult coding tasks or long multi-step reasoning, Gemini 3.6 Flash is the better fit.
Top comments (0)