Google Launches Gemini 3.7 Flash for Coding and AI Agents

Google said on the 14th that it had unveiled its generative AI model, "Gemini 3.7 Flash." It is being offered to developers, enterprise customers, and individual users. The Flash lineup is the Gemini family’s speed- and cost-efficiency-focused product tier.

Google said the new model is aimed at coding and AI agent tasks. The key improvements in "Gemini 3.7 Flash" are its ability to handle coding and agent-based, multi-step tasks.

The launch comes about 3 weeks after the release of "Gemini 3.6 Flash." The follow-up release is intended to strengthen software development and work automation performance.

According to Google’s published benchmark results, "Gemini 3.7 Flash" scored 43.6% on "FrontierCode 1.1 Main," higher than "Gemini 3.6 Flash"’s 34.4% on the same item. It also scored 65.3% on "DeepSWE v1.1," outperforming "Gemini 3.6 Flash"’s 49.0% on the same item.

Google said it demonstrated performance gains by posting higher figures than the previous model or 3.6 Flash across several evaluations, including web development, document processing, and work automation. In connection with improved web development performance, it said it scored an Elo rating of 1588 in a user-preference comparison metric on "WebDev Arena," versus 3.6 Flash’s 1538. It also highlighted improved UI implementation capabilities based on screenshot, image, and design-system inputs.

It also said document-processing and work-automation metrics improved. In a complex document-processing evaluation on "GDP.pdf," it posted 34.0%, beating 3.6 Flash’s 22.0%. In "AutomationBench," an evaluation of real-world business task execution, it posted 30.4%, up from the previous model’s 17.0%.

Google also cited improvements in multi-step planning for agent tasks, tool-calling capabilities, and instruction-following abilities. Based on this, it said the goal is to reduce the number of times developers need to intervene directly and to cut down on repeated requests for the same task.

Google will apply promotional pricing to Gemini 3.7 Flash through the end of the year. The price is USD 0.75 per 1 million input tokens and USD 3.75 per 1 million output tokens. That pricing is about half of 3.6 Flash’s original price.

The same promotional pricing will also apply to 3.6 Flash during the same period. Then, starting on January 1, 2027, pricing for both models will change. Both models will be charged USD 1.50 per 1 million input tokens and USD 7.50 per 1 million output tokens.

Gemini 3.7 Flash is available in Google AI Studio, the Gemini API, Android Studio, and "Google Antigravity." "Google Antigravity" is Google’s agent development environment. The enterprise distribution channels are the "Gemini Enterprise Agent Platform" and "Gemini Enterprise." Starting today, 3.7 Flash is also being applied to the personal AI agent "Gemini Spark." "Gemini Spark" is an agent service that continuously carries out tasks according to user instructions, and it is available to Google AI Pro and Ultra subscribers in more than 160 countries.

Google said this update improved Spark’s ability to use Google Workspace apps, and that the accuracy and output quality of tasks requiring multiple functions to be used in sequence also improved. As a result, it said Spark can now continuously carry out multi-step work such as gathering files, drafting emails, and updating status documents in a single flow.

Source: IT DAILY · Kim Byung-joo
Original: https://www.itdaily.kr/news/articleView.html?idxno=241003

References

• https://www.itdaily.kr/news/articleView.html?idxno=241003