Google releases Gemini 3 Flash

What happened

On 17 December 2025, Google released Gemini 3 Flash, describing it as frontier intelligence built for speed^[1]. The model did not stay in a preview corner: it became the default model in the Gemini app, taking the place of 2.5 Flash^[1]. Google also made it available to everyone through the Gemini app and through AI Mode in Search^[1].

For developers and companies, the release came in preview through the Gemini API in Google AI Studio, and to enterprises through Vertex AI and Gemini Enterprise^[1].

The published numbers

Google reported two benchmark results for the model: 90.4% on GPQA Diamond and 78% on SWE-bench Verified^[1].

Both are standard ways of testing a model, and each measures a different kind of work. GPQA Diamond is a set of hard science questions written to be difficult even for people with advanced training in the field; a high score suggests the model can reason through technical material rather than recall a well-known answer. SWE-bench Verified is a software-engineering test built from real coding tasks drawn from open-source projects; a model has to read the problem, work with an existing codebase, and produce a change that passes the project's tests. Publishing both means Google is claiming strength in scientific reasoning and in practical coding at the same time.

Price and speed

Google priced Gemini 3 Flash at $0.50 per million input tokens and $3 per million output tokens^[1]. Tokens are the small pieces of text — roughly word fragments — that a model reads and writes, and API pricing is quoted per million of them so that developers can estimate the cost of a workload.

On speed, Google said the model was about 3 times faster than 2.5 Pro^[1]. The company added that this comparison was based on Artificial Analysis benchmarking, a third-party evaluation service that measures how quickly models respond^[1]. Naming the source matters because speed depends heavily on how a test is run.

Why it matters

Within a model family, the Flash tier is the faster, cheaper option, while the Pro tier is the larger and more expensive one. Making a Flash model the default in the Gemini app means that the version most people meet first is the quick, low-cost one rather than the most capable one. Combined with a published benchmark pair and a per-token price, the release set out a specific position: a model that is meant to be used constantly, in search results and in a consumer app, rather than only through an API for occasional hard problems.