What is Gemini 3.5 Flash-Lite?

Google refreshed its Flash tier on July 21, 2026, and Gemini 3.5 Flash-Lite is the budget end of it. It’s the cheapest, fastest Gemini you can call today, built for high-volume, latency-sensitive work: classification, extraction, short chat replies, and simple retrieval answers where throughput and cost matter more than deep reasoning. If you’re firing millions of small requests a day, this is the tier Google wants in your hot path.

The refresh shipped three models at once, and the naming trips people up, so let’s clear that up fast and then get into the specs, pricing, benchmarks, and how to test the endpoint yourself.

What is Gemini 3.5 Flash-Lite?

Gemini 3.5 Flash-Lite is the smallest, cheapest model in Google’s Gemini family. It’s tuned for speed and volume, not for the hardest reasoning. Google clocks it at roughly 350 output tokens per second, so responses feel near-instant even under load. You call it through the Gemini API with the model id gemini-3.5-flash-lite.

Think of it as the tier you point at boring, repetitive, high-frequency jobs. Tagging support tickets. Pulling fields out of a document. Answering a short question from a chunk of retrieved text. Powering a chat widget that needs to reply before the user notices a delay. In each of those, you’re sending a lot of requests and you want each one cheap and quick. Flash-Lite is Google’s answer for that shape of workload.

It landed as part of a three-model Flash refresh: the new workhorse Gemini 3.6 Flash, this Gemini 3.5 Flash-Lite, and a gated security model called Gemini 3.5 Flash Cyber. You can read Google’s own rundown on the Google blog and the model details on the DeepMind Flash page. Flash-Lite sits at the bottom of that stack on price and at the top on raw speed.

Wait, why is it 3.5 and not 3.6?

Here’s the confusing part. The new workhorse model is Gemini 3.6 Flash. But Flash-Lite stayed on 3.5, and so did the gated Cyber security model. So one launch, one day, three models, two version numbers. That’s not a typo on Google’s part or a mistake in this article. Google shipped mixed versioning: the flagship Flash jumped to 3.6, while the Lite and Cyber variants kept the 3.5 label.

Why does Google do this? The version number tracks the model, not the launch event. The workhorse got a bigger generational bump, so it earned the 3.6 tag. Flash-Lite is a fresh model too, but Google chose to keep it aligned with the 3.5 line. Annoying to track, but that’s how the labels landed.

One more source of confusion worth killing now. Gemini 3.5 Flash-Lite is not the older Gemini 3.1 Flash-Lite from the previous generation. Same “Flash-Lite” name, different model, different numbers. If you built on 3.1 Flash-Lite earlier, this is the newer, faster replacement, not a rename of what you already had. Check the model id before you assume: gemini-3.5-flash-lite is the new one.

Specs and pricing

Here’s what Flash-Lite costs and how it behaves. All prices are per million tokens through the Gemini API.

That $0.30 input and $2.50 output is the cheapest published rate in the current Gemini lineup. For comparison, the 3.6 Flash workhorse runs $1.50 input and $7.50 output, so Flash-Lite is roughly a fifth of the input cost and a third of the output cost. When you’re processing millions of short calls, that gap is the whole point. You can confirm current numbers on the Gemini API pricing docs, since Google updates that page directly.

Context caching cuts the cost further for prompts you send repeatedly, like a fixed system prompt or a long reference document. You pay a small rate to cache the tokens plus an hourly storage fee, and cached input is much cheaper on each reuse.

How well does it perform?

The headline number: Gemini 3.5 Flash-Lite scores 54% on Terminal-Bench 2.1, up from 31% on its predecessor. That’s a real jump for a Lite-class model, and it means the small tier is now genuinely useful for tasks that used to slip through it.

Where it shines: classification, extraction, chat responses, and simple retrieval-augmented answers. Those are the jobs where a fast, cheap model earns its keep. Sorting inputs into buckets, pulling structured data out of messy text, answering a short question grounded in a retrieved passage, keeping a lightweight assistant responsive. Flash-Lite handles all of that well and fast.

Now the honest limit. Flash-Lite trades quality for cost and speed. It is not the model for the hardest agentic coding, long multi-step tool chains, or deep reasoning problems. When a task needs to plan across many steps, call tools reliably in sequence, or reason through a tricky codebase, you’ll want Gemini 3.6 Flash or a larger model instead. Picking Flash-Lite for that work saves money right up until the wrong answers cost you more than you saved. Match the tier to the task.

Where Google uses Flash-Lite

Google isn’t just selling Flash-Lite to developers. It’s rolling the model into Google Search. That’s a useful signal: when Google puts a model in front of Search-scale traffic, it’s betting the model is fast enough and cheap enough to run on an enormous number of queries without falling over on latency or budget.

For you, that’s a proof point. The same properties that make Flash-Lite fit Search, high throughput and low cost per call, are exactly what make it fit a busy production endpoint. If it can serve Search traffic, it can serve your classification pipeline.

Flash-Lite vs Gemini 3.6 Flash: which should you pick?

Short version: pick Flash-Lite when the job is simple, high-volume, and speed-sensitive. Pick 3.6 Flash when the job needs stronger reasoning, coding, or multi-step agent work.

Flash-Lite wins on price and latency. It’s the cheapest Gemini and runs at about 350 tokens per second, so it’s the right default for classification, extraction, short chat, and simple RAG at scale. Gemini 3.6 Flash costs more per token but reasons harder, codes better, and holds up on agentic workflows where Flash-Lite would stumble. The 3.6 Flash pricing reflects that: you pay for the extra capability.

A practical pattern is to route by difficulty. Send the easy, frequent calls to Flash-Lite and escalate the hard ones to 3.6 Flash. You get cheap throughput on the bulk of traffic and stronger reasoning only where it earns its cost. For a full side-by-side on speed, price, and quality, see Gemini 3.5 Flash-Lite vs 3.6 Flash.

How to access and test it

Flash-Lite runs on the Gemini API. The fastest way to start is Google AI Studio: grab an API key, and the free tier lets you try the model before you wire up billing. Note that the free tier is rate-limited, and Google may use free-tier data to improve its products, so keep sensitive inputs on a paid key. The full reference lives at the Gemini API docs. Set the model to gemini-3.5-flash-lite and you’re calling the cheapest tier.

Once your requests work, you want to keep them working. Prices, rate limits, and response shapes shift as Google updates the API, and a Lite model that’s fast but occasionally wrong needs assertions to catch drift. That’s where an API client helps. With Apidog you can build the POST request to the Gemini endpoint, store your API key as an environment variable so it never lives in the request body, and add assertions on the status code and the JSON fields you depend on.

From there you can turn that request into a saved regression test and schedule it to run so you catch a broken response or a pricing change before your users do. Apidog doesn’t run the model or replace the Gemini API; it’s the client you use to call, validate, and monitor the endpoint. Download Apidog if you want to test the gemini-3.5-flash-lite endpoint alongside the rest of your API stack.

FAQ

Is Gemini 3.5 Flash-Lite the same as Gemini 3.6 Flash? No. They’re two different models from the same July 21, 2026 refresh. Flash-Lite is the cheapest, fastest tier for simple high-volume work. 3.6 Flash is the pricier workhorse with stronger reasoning and coding.

Why is it 3.5 and not 3.6? Mixed versioning. Google bumped the workhorse Flash to 3.6, but kept Flash-Lite and the gated Cyber model on the 3.5 label in the same launch. The number tracks 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. The older 3.1 Flash-Lite is the previous generation. Same family name, different model and version, so check the model id gemini-3.5-flash-lite to be sure you’re on the new one.

Is Gemini 3.5 Flash-Lite free? There’s a free tier through Google AI Studio, and it’s rate-limited. It’s fine for testing and light use. For production volume you move to a paid API key, and Google may use free-tier data to improve its products.

What is Flash-Lite best at? Classification, extraction, short chat replies, and simple retrieval-augmented answers at high volume. It’s not the pick for hard agentic coding or long multi-step reasoning, where 3.6 Flash fits better.