What happened?
On August 13, 2026, Google DeepMind announced Gemini 3.7 Flash, its new model built for coding and AI agents. The company said it is the most capable working model in the Flash series. The announcement came just three weeks after the release of Gemini 3.6 Flash, which Google described as a direct result of developer feedback and algorithmic innovations.
The new model shows marked improvement over its predecessor in tasks like code debugging and problem-solving. On the FrontierCode 1.1 Main benchmark, Gemini 3.7 Flash scored 43.6 percent compared to Gemini 3.6 Flash's 34.4 percent. On DeepSWE v1.1, 3.7 Flash scored 65.3 percent, outperforming 3.6 Flash's 49.0 percent.
In web development, the model can produce more functional interfaces and complete applications with fewer prompts. On Arena.ai's WebDev Arena benchmark, Gemini 3.7 Flash scored 1588 Elo, surpassing Gemini 3.6 Flash's 1538. In knowledge-intensive fields such as finance, law and biosciences, the model also demonstrates stronger reasoning and accuracy; on the GDP.pdf evaluation it scored 34.0 percent versus the previous model's 22.0 percent. On AutomationBench, it achieved a 30.4 percent success rate in completing real-world workflows, up from 3.6 Flash's 17.0 percent.
Why it matters
Google is offering Gemini 3.7 Flash at an introductory price valid through the end of the year: $0.75 per 1 million input tokens and $3.75 per 1 million output tokens. This is half the original per-token price of Gemini 3.6 Flash. According to Google, this pricing allows developers and enterprise customers to scale production-ready AI agents at lower cost.
The model also performs more consistently in multi-step planning and tool-calling tasks, reducing the need for manual oversight and retries in engineering workflows. Google said early customer feedback indicated the model delivers significantly better results at lower cost compared to 3.6 Flash in terms of both performance and precision.
Google noted that Gemini 3.7 Flash launches with updated safety measures against chemical, biological, radiological and nuclear (CBRN) misuse as well as cyberattack abuse. The company described this as part of its ongoing work to expand and strengthen its Frontier Safety measures.
What's next
Gemini Spark, the personal AI agent available to Google AI Pro and Ultra subscribers in more than 160 countries, began running on Gemini 3.7 Flash as of August 13, 2026. Google said this update makes Spark more effective for knowledge work tasks by enabling more efficient tool use with Google Workspace apps.
Developers can access the model through Google Antigravity, the Gemini API, Google AI Studio and Android Studio. Enterprise users can reach it via the Gemini Enterprise Agent Platform and the Gemini Enterprise app, while individual users in supported countries can use the model through Spark in the Gemini app.