Gemini 3.5 Flash-Lite: what it is, price, and who it's for

What Gemini 3.5 Flash-Lite actually is

Google splits its Gemini family into two lanes. Pro models are the top tier for hard reasoning. Flash models are the workhorses: cheaper, faster, tuned for apps that make thousands of calls a day. Flash-Lite is the lightest of those workhorses, positioned by Google for high-volume agentic tasks, translation, and simple data processing.

Think of it as the model you point at work where the individual task is easy but the volume is brutal: classifying a support ticket, extracting fields from a PDF, translating a chat message, running a first-pass search. It handles text, image, video, audio, and PDF input, supports function calling, structured outputs, code execution, caching, and search grounding, and it carries the same ~1M-token context window as its bigger siblings (about 1,500 A4 pages of input).

What it deliberately leaves out tells you who it's for. Flash-Lite does not support computer use, image generation, audio generation, or the Live API. If your agent needs to drive a browser or generate media, that's Gemini 3.6 Flash or a Pro model, not this one. Flash-Lite stays narrow on purpose: read text, decide, write text, do it fast and cheap.

Gemini 3.5 Flash-Lite pricing

Here's the full picture. This is the reason the model exists, so it's worth reading closely before you plug it into a spreadsheet.

A few things worth flagging:

• The free tier has a catch. On the free tier Google uses your content to improve its products; on any paid tier it doesn't. For anything touching customer data, that alone rules out the free tier.
• Batch mode is a flat 50% off, which is the tier most high-volume jobs should actually run on. Same for flex inference. If the work doesn't need to be real-time, you're paying $0.15/$1.25.
• Cache hits are cheap. A cached input token costs $0.03 per 1M, a 90% cut, so repeated context (your system prompt, a knowledge snippet) barely registers.
• Search grounding is metered separately. You get 5,000 grounded prompts a month free across the Gemini 3 family, then $14 per 1,000 search queries.

One honest note from Artificial Analysis: at $0.30/$2.50 it's ranked #87 of 152 on raw price, so "Lite" doesn't mean "cheapest on the market." It means cheapest in the Gemini 3.5 line while still posting a top-tier intelligence score. For how this sits against the consumer Gemini plans and the other API tiers, our Gemini pricing guide has the full breakdown and Gemini Workspace pricing covers the business plans.

Speed is the headline feature

If pricing is why you'd shortlist Flash-Lite, speed is why you'd keep it. Artificial Analysis clocked it at about 490 tokens/second, ranked #2 of 152 models tested; Google cites 350 tokens/second on its own harness. Either number puts it near the top of the field.

There's one asterisk worth knowing before you promise a product manager sub-second replies. The time-to-first-token sits around 6 seconds median, on the high end of its tier, because the model's thinking step runs before the first visible token. So the stream is blisteringly fast once it starts, but the start can lag on a hard prompt. For most support work (short answers, simple classification) that thinking budget is small and the lag is negligible; for anything you're timing to the millisecond, test it on your own prompts first.

Against its own predecessor, the generation jump is clean across agentic coding and knowledge-work benchmarks:

Flash-Lite or 3.6 Flash? Pick the right one

This is the decision most teams get wrong, because the two shipped on the same day and the names barely differ. They're built for opposite ends of the same job. Walk the picker below.

The rule of thumb: reach for Flash-Lite when each item is simple and the volume is the challenge, like first-pass ticket triage, tier-1 deflection, or routine live-chat deflection. Reach for 3.6 Flash when the work really needs reasoning across several systems. Whether you build a real AI agent or a scripted bot is a separate decision from the model one.

The other models that shipped that day

Flash-Lite didn't launch alone. Google shipped three models on July 21 and held one back, and knowing the lineup saves you from picking the wrong tier.

Gemini 3.6 Flash is the new headline workhorse, at $1.50 input / $7.50 output with built-in computer use and about 17% fewer output tokens than 3.5 Flash. Gemini 3.5 Flash Cyber is a security specialist fine-tuned to find and fix vulnerabilities inside Google's CodeMender agent, and it's available only to governments and trusted partners in a limited pilot.

The notable absence: the flagship Gemini 3.5 Pro. Bloomberg reported it hit internal delays after missing its own performance goals, and DeepMind's Logan Kilpatrick said it's still testing with partners. So Google's move here is telling: keep the cheap, high-volume tier fresh while the flagship cooks. If you need a top-tier model today, our Gemini alternatives roundup is the honest place to look, Gemini vs Claude and Gemini vs ChatGPT cover the head-to-heads, and AI customer service companies maps the vendors building on top of these models.

What this means if you're building AI for support

Here's where I have to be straight with you, because this is the part every model-launch post skips for support teams.

A faster, cheaper model is good news. But swapping in Flash-Lite does not give you a working support agent, any more than a faster CPU gives you an app. I spend my days wiring integrations and APIs into eesel, and across three-plus years of putting AI on live support queues, the failure we've watched over and over isn't the model being dumb, it's a confident-sounding bot giving a wrong answer to a real customer because nobody built the layers around the model. The model is the bottom of the stack, not the whole thing.

Those layers are the actual work. Retrieval, so the model answers from your knowledge base and past tickets instead of guessing. Guardrails and scope, so it hands off cleanly instead of inventing a refund policy. Integration, so it can act inside your ticketing system. And testing, so you know how it behaves before it touches a customer, not after. This is the gap between a raw API and real AI customer service software, and it's why AI chatbot problems almost always trace back to the plumbing, not the model.

It's also why I'd push back on wiring the raw Gemini API into your helpdesk yourself. You'd be rebuilding retrieval, hallucination guardrails, simulation, and every helpdesk integration from scratch, then maintaining it as Google ships a new model every few weeks (three in one day, this time). That churn is the whole case for purpose-built helpdesk AI over a DIY stack: the model tier is a swappable line item, the surrounding system is the product.

Try eesel

This is exactly the layer eesel is built to be. You plug it into your existing helpdesk, it learns from your past tickets and knowledge base out of the box, and it runs on top of frontier models like Gemini so you get the speed and cost savings of a model like Flash-Lite without wiring the API yourself.

The differentiator that matters most here is the one Flash-Lite can't give you on its own: eesel lets you simulate the agent against your historical tickets before it ever answers a live customer, so you see the resolution rate and the exact replies it would have sent. And because it's priced on outcomes, not per token, you skip the token math on every ticket. If you're evaluating models to build support automation, start from the outcome you want and let the platform pick the plumbing. It's free to try.