Gemini 3.7 flash model released for coding and agents

Google has announced the release of Gemini 3.7 Flash, positioned as its most intelligent workhorse model to date for coding and multi-agent systems. The model arrives just three weeks after the introduction of Gemini 3.6 Flash, incorporating direct developer feedback and new algorithmic improvements.

Software engineering and web development

The new model demonstrates notable gains in software engineering tasks, including debugging and issue resolution. Benchmark evaluations show improvements in first-pass code accuracy and the generation of production-ready code. Specifically, the model scored higher on FrontierCode 1.1 Main at 43.6% compared to the previous version's 34.4%, and reached 65.3% on DeepSWE v1.1.

In web development workflows, Gemini 3.7 Flash produces functional layouts and feature-complete applications with fewer prompts. It achieved an Elo score of 1588 on Arena.ai's WebDev Arena, outperforming the prior score of 1538.

Knowledge work and pricing

For knowledge-dense fields such as law, finance, and biosciences, the model provides improved reasoning capabilities. On the GDP.pdf document-processing benchmark, it scored 34.0%, up from 22.0% in the previous iteration.

Gemini 3.7 Flash is available at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens through the end of the year. This pricing structure is intended to help developers scale production-ready agents more cost-effectively.

Ecosystem integration and safety

Gemini Spark, the personal AI agent available to subscribers in over 160 countries, has integrated the new model to improve tool use for workspace applications and multi-skill workflows.

The model ships with updated safeguards designed to prevent misuse in specialized domains such as cyber offense and chemical, biological, radiological, and nuclear applications, while supporting authorized use cases.