Gemini 2.5 Flash-Lite

What it is

Gemini 2.5 Flash-Lite is the lowest-cost member of Google's Gemini 2.5 model family. Google previewed it on 17 June 2025 and positioned it as an upgrade over the 1.5 and 2.0 Flash models for high-throughput work, naming classification and summarisation as the typical jobs.^[1] High throughput here means the model is called a very large number of times with relatively small inputs: sorting support tickets into categories, tagging documents, or condensing a long text into a few lines. In workloads like these the price of each call and the speed of the answer matter more than the hardest reasoning, so the cheapest model that is good enough usually wins.

Thinking off by default

The Gemini 2.5 family was built around "thinking": the model writes out intermediate reasoning before it answers, which costs extra tokens and extra time. Flash-Lite is the exception. Unlike the other 2.5 models, thinking is off by default, and developers can set a controllable thinking budget when they want it.^[1] Google described the model as giving developers an even lower-cost option, with or without thinking.^[1] A thinking budget is a cap on how much internal reasoning the model may spend before answering, so a team can keep most traffic on the cheap path and still allow a longer deliberation for the requests that need it.

What it can do

Besides plain text generation, the model supports Grounding with Google Search, code execution, URL context and function calling.^[1] Grounding with Google Search lets it pull in current results from the web instead of relying only on what it learned during training. Code execution lets it run code as part of producing an answer. URL context lets it read the contents of a page the developer supplies. Function calling lets it emit a structured request for an outside tool, which the surrounding application then carries out. Together these are the plumbing features that turn a model from a text generator into a component of a working application.

Pricing

Google listed input at $0.10 per 1 million tokens for text, image and video, and output at $0.40 per 1 million tokens.^[3] Models are billed per token, the small chunks of text (or patches of image, frames of video) that the model reads and writes, so these figures describe the cost of volume rather than the cost of one request. Output is priced higher than input because generating tokens is the more expensive direction of the work. For the kind of application Flash-Lite was aimed at, where a single document might be sent through the model alongside thousands of others, the per-million figures are the ones that decide whether a product is affordable.

Timeline

It became generally available on 22 July 2025 under the API name gemini-2.5-flash-lite.^[2] An updated preview was released on 25 September 2025 and retired on 31 March 2026.^[2] In 2026 it was succeeded by Gemini 3.1 Flash-Lite.^[2]

Where it fits

The Flash-Lite line sits at the bottom of the price list, and that position is less marginal than it sounds. In a working application, most calls are routine: a short question, a label, a summary. The cheapest tier is where the bulk of production traffic tends to go, while the larger models are reserved for the requests that actually need them. Making thinking optional rather than fixed reflects the same idea, that reasoning effort is something a developer should be able to dial up or down per request instead of paying for it everywhere. Flash-Lite is the version of that idea taken to its cheapest end, and the later Gemini 3.1 Flash-Lite continued the naming in the same slot.^[2]