Gemini 3.5: Google Revolutionizes AI with Ultra-Fast Agents

Gemini 3.5: A Major Advancement in Artificial Intelligence

Google has recently unveiled Gemini 3.5, a new series of artificial intelligence models that combines cutting-edge intelligence with action capability. This innovation marks a significant turning point in the development of more efficient intelligent agents. The launch of this series begins with 3.5 Flash, a model that stands out for its exceptional performance in handling complex and prolonged tasks, providing tangible utility in the real world. Starting today, 3.5 Flash is accessible to billions of users around the globe:

• Available to everyone via the Gemini app and Google Search's AI mode.
• Accessible to developers on the Google Antigravity platform, as well as through the Gemini API in Google AI Studio and Android Studio.
• Offered to businesses through the Gemini Enterprise Agent and Gemini Enterprise platforms.

Meanwhile, Google is working on 3.5 Pro, currently used internally, with deployment expected next month.

3.5 Flash: Unmatched Performance for Agents and Coding

The Gemini 3.5 Flash model positions itself as a benchmark in artificial intelligence, competing with the most advanced models in several aspects while offering the characteristic speed of the Flash series. This model outperforms Gemini 3.1 Pro on demanding benchmarks such as Terminal-Bench 2.1 with a score of 76.2%, GDPval-AA with an Elo of 1656, and MCP Atlas with 83.6%. It also excels in multimodal understanding with a score of 84.2% on CharXiv Reasoning. In terms of speed, 3.5 Flash is four times faster than other leading models in terms of output tokens per second.

Ranked in the upper right quadrant of the Artificial Analysis index, 3.5 Flash demonstrates advanced intelligence combined with exceptional speed, proving that it is no longer necessary to choose between quality and speed.

3.5 Flash: Large-Scale Agentic Tasks

The balance between speed and performance makes 3.5 Flash an ideal tool for complex and prolonged agentic tasks. What previously took days for a developer or weeks for an auditor can now be accomplished in record time, often at less than half the cost of competing models. 3.5 Flash is capable of planning, building, and iterating quickly to solve real-world problems, whether it's developing new applications, maintaining codebases, or preparing financial documents.

Paired with the updated Antigravity tool, 3.5 Flash becomes a powerful engine for deploying collaborative sub-agents capable of solving large-scale problems for the most demanding use cases. Under supervision, it can reliably execute multi-step workflows and coding tasks while maintaining peak performance.

3.5 Flash: Tangible Impact in the Real World

The agentic capabilities of 3.5 Flash are already having a significant impact on developers and businesses. In developing the 3.5 series, Google closely collaborated with industry partners to identify points of complexity and effort in their workflows. The results are compelling:

• Shopify uses parallel sub-agents to analyze complex data over extended periods, enabling more accurate growth forecasts for merchants on a global scale.
• Macquarie Bank is exploring how 3.5 Flash can accelerate client onboarding by analyzing complex documents over 100 pages long, extracting relevant information, and formulating reliable recommendations with low latency.
• Salesforce integrates 3.5 Flash into Agentforce to reliably automate complicated business tasks, deploying multiple sub-agents that maintain context and execute complex multi-step tool calls.
• Ramp uses 3.5 Flash to provide smarter and more reliable OCR through multimodal understanding of complex invoices, combined with reasoning on historical patterns.
• Xero deploys agents to autonomously manage complex workflows over several weeks, such as identifying suppliers and gathering information for 1099 tax forms, allowing small businesses to automate tedious administrative tasks.
• Databricks employs agentic workflows to monitor and retrieve real-time information, analyzing massive datasets to diagnose issues, identify solutions, and propose recommendations for data scientists.

Personal AI Agents: The Era of 3.5 Flash

3.5 Flash is now the default model for the Gemini app and AI mode in Search worldwide. At the I/O event, Google demonstrated how its agentic capabilities power new features to integrate cutting-edge intelligence into users' daily lives.

The new personal AI agent, Gemini Spark, is built on 3.5 Flash. Operating continuously, it helps navigate users' digital lives, acting on their behalf while remaining under their direction. Gemini Spark is currently being deployed to trusted testers, with a beta version expected for Google AI Ultra subscribers in the United States next week.

Gemini 3.5: Enhanced Security

The development of Gemini 3.5 has been carried out in accordance with Google's Frontier Safety Framework. Protections against cyber threats and CBRN risks have been strengthened, thereby reducing the risks of generating harmful content or erroneously refusing to respond to safe queries. These improvements are the result of new advanced security training and mitigation measures, including interpretability tools that help verify and understand the AI's internal reasoning before it provides a response.

Immediate Availability of 3.5 Flash

Gemini 3.5 Flash is now available via Google Antigravity, the Gemini API in Google AI Studio and Android Studio, as well as on the Gemini Enterprise Agent and Gemini Enterprise platforms. It is also available to everyone in the Gemini app and the AI mode of Search. The entire Gemini team is eager to see the creations that this powerful tool will enable.