Google unveils Gemini 3.5 Flash with frontier AI capabilities for complex agent workflows

Google has introduced Gemini 3.5, its latest AI model family that combines advanced intelligence with autonomous action capabilities. The release represents a significant step toward building more capable AI agents that can execute complex, multi-step tasks independently.

The company is launching the series with 3.5 Flash, a model that delivers frontier-level performance while maintaining the speed users expect from the Flash lineup. The model is designed to excel at coding and complex long-term tasks that provide real-world utility, marking a shift toward AI systems that can take meaningful action rather than just provide information.

Performance benchmarks and availability

Gemini 3.5 Flash delivers intelligence comparable to large flagship models while operating at speeds four times faster than other frontier models. The model shows impressive performance across several challenging benchmarks:

• Terminal-Bench 2.1: 76.2%
• GDPval-AA: 1656 Elo rating
• MCP Atlas: 83.6%
• CharXiv Reasoning: 84.2% for multimodal understanding

The model is now available globally through multiple channels including the Gemini app, AI Mode in Google Search, Google Antigravity development platform, and enterprise solutions. A more advanced 3.5 Pro model is currently being tested internally and will launch next month.

Agent workflows solve real-world problems

The model’s ability to handle long-term autonomous tasks addresses a key limitation in current AI systems. Where developers previously spent days or auditors spent weeks on complex tasks, 3.5 Flash can complete similar work in a fraction of the time at less than half the cost of competing models.

When combined with Google’s Antigravity platform, the model can deploy multiple collaborative sub-agents to tackle large-scale problems. Examples include automatically organizing unstructured assets, synthesizing research papers while coding functional games, and modernizing legacy codebases to current frameworks like Next.js.

The model also generates more interactive and visually rich outputs, creating everything from animated research presentations to complete branding concepts and user interface designs within minutes.

Enterprise adoption and real-world impact

Several major companies are already implementing 3.5 Flash for complex business processes:

• Shopify uses parallel sub-agents for global merchant growth forecasting
• Macquarie Bank pilots customer onboarding automation with 100+ page document analysis
• Salesforce integrates the model into Agentforce for enterprise task automation
• Ramp improves invoice processing through multimodal understanding
• Xero automates multi-week workflows like supplier identification for tax forms
• Databricks monitors datasets and diagnoses issues for data scientists

This enterprise adoption suggests the technology is moving beyond experimental use cases into production environments where reliability and accuracy are critical.

Personal AI agent introduction

Google is launching Gemini Spark, a personal AI agent powered by 3.5 Flash that runs continuously to help users manage their digital lives. The agent can take actions on behalf of users while remaining under their control and direction.

Gemini Spark begins rolling out to trusted testers immediately, with beta access planned for Google AI Ultra subscribers in the US next week. The agent represents Google’s vision for AI that actively helps users rather than simply responding to queries.

Safety measures and future implications

The model was developed under Google’s Frontier Safety Framework with enhanced cyber and CBRN (chemical, biological, radiological, nuclear) safeguards. The company has implemented advanced safety training and interpretability tools that analyze the AI’s reasoning process before generating responses.

The launch of Gemini 3.5 Flash signals a significant shift in AI development toward autonomous agents capable of executing complex, real-world tasks. As these systems become more capable of independent action, they could reshape how businesses operate and how individuals interact with technology in their daily lives.