Gemini 3 Flash: Google's fast model becomes the default and targets coding work

On December 17, Google launched Gemini 3 Flash, the fast and cheap member of the Gemini 3 family. “Flash” models are usually treated as the budget option, the one you reach for when money is tight. The announcement tries to change that conversation: Google claims 3 Flash beats Gemini 2.5 Pro while being much faster, and makes it the default model in the Gemini app and in Search. For developers, what matters is price, availability, and how far the numbers can be trusted.

Pricing and where it is available

The published prices, per million tokens:

input: $0.50;
output: $3;
audio input: $1.

For developers, the model is in preview in the Gemini API, Google AI Studio, Gemini CLI, Google Antigravity and Android Studio. For enterprises, it is in Vertex AI and Gemini Enterprise.

On the consumer side, Gemini 3 Flash replaces 2.5 Flash as the default model in the Gemini app and is also rolling out as the default in AI Mode in Search. That matters to developers because it means the model will be exercised at enormous scale right away: behavior problems tend to surface (and get fixed) quickly.

The numbers, according to Google

The benchmarks Google reports:

GPQA Diamond: 90.4%;
Humanity’s Last Exam: 33.7% without tools;
MMMU Pro: 81.2%, which Google describes as state of the art and comparable to Gemini 3 Pro;
SWE-bench Verified: 78%, a score that, according to the announcement, beats the 2.5 series and Gemini 3 Pro itself.

The SWE-bench figure is the one that stands out to me. A Flash-tier model beating the same generation’s Pro on a benchmark of fixing real bugs in real repositories is unusual, and Google clearly wants to position 3 Flash as a workhorse for coding agents.

On speed and running cost, Google says the model is 3x faster than 2.5 Pro, based on Artificial Analysis benchmarking, and uses 30% fewer tokens on average than 2.5 Pro on typical traffic. That second point matters as much as the per-token price: a model that “thinks” less to reach the same answer lowers both the final bill and perceived latency.

The announcement also says that since Gemini 3 launched, the API has been processing more than 1 trillion tokens per day, and it names customers such as JetBrains, Bridgewater Associates and Figma.

What this changes for developers

In practice, I see three immediate uses:

Coding agents and terminal tools. With day-one availability in Gemini CLI and Antigravity, plus the SWE-bench number Google published, 3 Flash is a natural candidate for agent loops that make many short calls, where latency and cost add up fast.
Replacing 2.5 models in production. If you have a pipeline running on 2.5 Flash or 2.5 Pro, an A/B test is worth it. Going by Google’s claims, you can gain quality over 2.5 Flash or cut cost and latency compared to 2.5 Pro.
Multimodal and audio. The separate audio input price, together with the MMMU Pro result, suggests Google sees the model as a base for applications that mix text, images and voice.

Caveats

A few things to keep in mind before migrating everything:

In the Gemini API the model is in preview. That usually means different rate limits and the possibility of changes before the stable release.
All of the numbers above come from Google itself. Public benchmarks rarely reflect your domain; run your own evaluations with real prompts.
“30% fewer tokens on average” depends on the kind of traffic. On long reasoning tasks, the savings could look quite different.

My take

Gemini 3 Flash is the Gemini 3 launch most likely to affect the day-to-day work of people building products, precisely because it is not the flagship. A low price, day-one presence in the CLI and IDEs, and a strong SWE-bench result make a combination that is hard to ignore. I would start by testing it as the default model for high-volume agent tasks, keeping bigger models for the cases where it fails consistently. What is still missing is seeing how it behaves outside the benchmarks Google chose, and only real-world use will show that.