Google DeepMind unveils Gemini 3.7 Flash — half the price of its predecessor

Google DeepMind has announced Gemini 3.7 Flash, a new release in its line of “affordable workhorse” models geared toward coding and agentic scenarios. It comes just three weeks after Gemini 3.6 Flash and, according to the company, is a direct result of developer feedback and algorithmic improvements.

The main draw is price: at launch, Gemini 3.7 Flash costs half as much as its predecessor — $0.75 per million input tokens and $3.75 per million output tokens. The introductory rate runs through the end of 2026; starting January 1, 2027, prices will double, to $1.50 and $7.50 respectively.

What's improved

Google points to a set of benchmarks on which the new model noticeably outpaces 3.6 Flash. In generating production-ready code, its score rose from 34.4% to 43.6% on FrontierCode 1.1 and from 49.0% to 65.3% on DeepSWE v1.1. In web development, the company claims, the model builds functional layouts from fewer prompts and scores 1588 Elo on WebDev Arena, against 1538 for its predecessor.

Knowledge-intensive fields — finance, law, the life sciences — get a separate mention. On GDP.pdf, a test of complex document processing, the model scored 34.0% versus 22.0%, and on AutomationBench, which measures performance on real-world business tasks, 30.4% versus 17.0%.

In its demos, Google showed how a text prompt gives rise to a playable 3D game (in tandem with Nano Banana), how a static PDF is assembled into an interactive report with live charts, and how a robot is trained in a three-agent loop with multimodal analysis.

Discipline over follow-up questions

Particular emphasis was placed on the model's “behavior” in agent workflows. According to DeepMind, 3.7 Flash handles obstacles better, clarifies the user's intent when necessary, and follows instructions more precisely, planning multi-step actions and tool calls more carefully. The upshot, the company promises, is less manual oversight and fewer retries.

The new model has already been put to work in Gemini Spark, a personal AI agent for Google AI Pro and Ultra subscribers, available in more than 160 countries. The update, Google says, improves work with Google Workspace apps: consolidating files, drafting emails, updating status documents.

Also highlighted are strengthened Frontier Safety safeguards against misuse in chemical, biological, radiological, and nuclear (CBRN) domains and against cyberattacks. The model is available to developers through the Gemini API in Google AI Studio and Android Studio, as well as in the Gemini Enterprise corporate platform.

Source: Google DeepMind