Google ships Gemini 3.7 Flash three weeks after 3.6, halves the price

Google has released Gemini 3.7 Flash, an update to its high-volume “workhorse” model line aimed at coding and agentic workflows, arriving just three weeks after 3.6 Flash. The headline pitch is economics as much as capability: an introductory price of $0.75 per million input tokens and $3.75 per million output tokens, roughly half what the previous version cost, held through the end of the year. Google frames the fast cadence as a payoff from developer feedback and algorithmic tweaks it plans to carry into later models.

The company backs the release with a spread of benchmark comparisons against 3.6 Flash: coding evals FrontierCode 1.1 (43.6% vs 34.4%) and DeepSWE v1.1 (65.3% vs 49.0%), a WebDev Arena Elo of 1588 versus 1538, plus gains on document-heavy and business-workflow tests like GDP.pdf and AutomationBench. Google also emphasizes softer improvements — better instruction-following, more deliberate multi-step planning, and cleaner tool calls — that it says cut down on retries and manual oversight in agent pipelines. As with most first-party model launches, these are vendor-reported numbers against the vendor’s own prior model, so the practical gap will depend on independent testing.

The model is going into production immediately behind Gemini Spark, Google’s always-on personal agent for AI Pro and Ultra subscribers, with tighter Google Workspace tool use for tasks like drafting emails and updating documents. It ships with updated Frontier Safety guardrails targeting CBRN and cyber-offense misuse, and is available to developers through the Gemini API and Antigravity and to enterprises via Google’s agent platform. The combination of a lower price and stronger agent behavior signals Google is competing hard on cost-per-token for teams running agents at scale.