Google announced Gemini 3.7 Flash on Thursday, three weeks after Gemini 3.6 Flash, which followed 3.5 Flash in July. Google described the new model as its most capable workhorse for coding and agents, built on developer feedback and algorithmic improvements rather than a full architectural overhaul.
The headline improvements are in software engineering. On DeepSWE v1.1, a long-horizon benchmark of how well models resolve real-world coding issues, 3.7 Flash scored 65.3 percent against 49.0 percent for 3.6 Flash, a jump of more than 16 points. On FrontierCode 1.1 Main, which evaluates production code quality, it improved from 34.4 to 43.6 percent. Google attributed the gains to better debugging and issue resolution, higher first-pass accuracy, and more disciplined multi-step planning.
Knowledge work also improved. The model roughly doubled its score on AutomationBench, a benchmark of real-world business workflows, from 17.0 to 30.4 percent, and rose from 22.0 to 34.0 percent on GDP.pdf, a complex document-processing test covering finance, law, and biosciences material. In web development, its WebDev Arena Elo rating rose from 1538 to 1588; Google said the model produces more functional layouts and feature-complete apps from fewer prompts, including design adherence based on reference screenshots, images, or design systems.
The pricing is notable. Google is offering introductory rates of $0.75 per million input tokens and $3.75 per million output tokens through the end of the year, half the launch price of 3.6 Flash, before standard rates of $1.50 and $7.50 take effect in January. That makes the model cheaper per token than its predecessor at launch.
Availability spans the Gemini API and AI Studio, Android Studio, Google Antigravity for agent-first workflows, and the Gemini Enterprise Agent Platform and app. Consumers on the Gemini app’s Spark tier, the personal agent Google introduced at I/O, get access starting Thursday for AI Pro and Ultra subscribers in more than 160 countries. Google cited improved tool use in Google Workspace and better accuracy on multi-step tasks.
The model ships with updated frontier safety safeguards aimed at misuse in the CBRN (chemical, biological, radiological, nuclear) and cyber offense domains, consistent with DeepMind’s bioresilience approach and its cyber program. Google published a model card with the details.
Sources: Introducing Gemini 3.7 Flash: our most intelligent workhorse model (Google, Aug 13, 2026); Gemini 3.7 Flash launches three weeks after last model, live in Spark (9to5Google, Aug 13, 2026); Google unveils Gemini 3.7 Flash AI model for coding and agent workflows (Reuters, Aug 13, 2026); Google announces Gemini 3.7 Flash just three weeks after previous release (Ars Technica, Aug 13, 2026)