Google has released Gemini 3.7 Flash, the latest update to its model lineup. This version focuses on software engineering and automated agent performance. The release arrives three weeks after the previous iteration and offers increased intelligence for complex coding tasks like debugging and issue resolution.
Benchmarks indicate measurable gains in code accuracy and production readiness. On the FrontierCode 1.1 Main test, the model achieved 43.6% compared to 34.4% in previous versions. Web development workflows also see improvements in layout generation and design adherence, scoring higher on industry-standard leaderboards.
Beyond coding, the model shows better reasoning capabilities in professional fields such as finance and law. Tests on complex document processing show performance jumps from 22.0% to 34.0% in specific benchmarks. It also demonstrates more effective execution of multi-step business workflows while reducing the need for manual oversight.
Developers can access the model through the Gemini API and Android Studio. The pricing structure is set at half the cost of the 3.6 Flash version, specifically $0.75 per million input tokens and $3.75 per million output tokens for the remainder of the year. This pricing model aims to make production-ready AI agents more accessible for scale.
Gemini Spark, the personal agent feature for Google AI Pro and Ultra subscribers, now runs on the 3.7 Flash model. The update enables better tool use within Google Workspace, assisting users with tasks like document management and drafting communications. Safety protocols against cyber misuse and other high-risk domains remain a priority in this release.