Google Ships Gemini 3.7 Flash at Half the Price of Its Predecessor

Google DeepMind released Gemini 3.7 Flash on 13 August 2026, positioning it as the workhorse model for coding and agent workloads that sits below the full Gemini 3 tier. The headline change is price. Gemini 3.7 Flash launches at roughly half the introductory cost of Gemini 3.6 Flash, the model it replaces, while posting stronger results on coding benchmarks.

That combination matters more than another leaderboard placing. Flash tier models are the ones doing the unglamorous, high volume work: autocomplete in an IDE, agent loops that call a tool a hundred times in a session, background jobs that summarise or classify. Halving the cost of that tier changes the maths for anyone running agents continuously rather than prompting one message at a time, which is increasingly how design and development tools use these models under the bonnet.

We use models at this tier constantly, wired into build pipelines and internal tooling rather than sitting in a chat window. A price cut here doesn't change what a single Gemini call can do for a client brief. It changes how liberally we can afford to use it, more agentic steps per task, more automated checking, more background processing that would have been too expensive to justify a year ago. The interesting story with Flash models has stopped being about raw capability and started being about what becomes affordable to run in the background of a normal working day.

The model card and pricing are live now for anyone building against the API, alongside the usual set of benchmark comparisons against the previous Flash generation. Google is positioning it specifically for coding and agent loops rather than as a general purpose replacement for the full Gemini 3 tier, which tracks with how most studios actually split their model usage: a capable, expensive model for the judgement calls, and a cheap, fast one doing the repetitive work underneath it.