en.Wedoany.com Reported - On May 19 local time, Google announced the launch of its new Gemini 3.5 model family at the 2026 I/O developer conference. As the first model in the series, Gemini 3.5 Flash was released first, featuring cutting-edge performance in agent operation and programming, particularly excelling at handling complex, long-duration tasks that generate practical application value. In addition, Google is currently accelerating the development of the more advanced Gemini 3.5 Pro model, which is already in internal use at Google and is expected to be officially launched to the market next month.
During the conference keynote, Google CEO Sundar Pichai stated that Gemini 3.5 Flash delivers cutting-edge capabilities at only half the cost, and sometimes less than a third, of comparable top-tier models, while maintaining a high level of intelligence. Pichai described Gemini 3.5 Flash as "extremely fast," and Google simultaneously announced that this model will become the default model for the Gemini app and Search AI Mode globally.
Gemini 3.5 Flash is defined by Google as its most capable agent and programming model to date. Its intelligence level has significantly closed the gap with large flagship models, while fully inheriting the Flash series' hallmark high-speed response. According to officially released data, in multiple key agent and coding benchmark tests, Gemini 3.5 Flash not only surpassed its own previous-generation flagship, Gemini 3.1 Pro, but also demonstrated extremely strong competitiveness.
A specific set of figures illustrates this performance leap more clearly. In the Terminal-Bench 2.1 test, it achieved a score of 76.2%, outperforming the previous flagship's 70.3%. In the GDPval-AA test, which measures comprehensive agent capability, it achieved an Elo score of 1656. In the MCP Atlas test, its score of 83.6% placed it at the top of the field. Similarly, in the multimodal understanding test CharXiv Reasoning, a score of 84.2% also made it the leader in that category.
Beyond performance gains, speed is another highly potent weapon of this model. The output speed of Gemini 3.5 Flash exceeds 280 tokens per second, a figure that is directly four times that of GPT-5.5 and Claude Opus 4.7. Google specifically emphasized that it is precisely this balance between ultra-high performance and extreme speed that makes it exceptionally adept at handling agent tasks that are long in duration and involve numerous steps.
Google stated in its official blog: "Users no longer need to choose between quality and response speed." The company also strengthened the cybersecurity protections of Gemini 3.5 Flash, making it "less likely to generate harmful content and less prone to refusing to answer on safety grounds." As for the heavyweight version, Gemini 3.5 Pro, it is currently for internal use only and is expected to be opened to a wider user base next month.
In the realm of AI agents, Google introduced Gemini Spark, a new general-purpose AI agent within the Gemini app capable of cross-platform reasoning over information in connected applications. Google stated that this tool can manage users' digital lives and perform actions on their behalf. Gemini Spark is currently in the testing phase and will first be available to trusted test users and Google AI Ultra subscribers next week.
Pichai said: "We have been bringing agents to developers and enterprises for some time now. Now, we are focused on bringing this cutting-edge capability to consumers safely and reliably, so everyone can benefit." As more internet users adopt chatbots, Google is working to convince traditional search users to trust it with tasks executed with minimal input. Amid a significant rise in the company's capital expenditure, Wall Street is watching whether Google can achieve deep product integration, and agent capabilities could become an important breakthrough point.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
