Google Introduces Gemini 3.5 Flash at I/O 2026: A Faster and Cheaper Model for AI Agents and Coding
Google has launched Gemini 3.5 Flash, a new AI model designed for faster and more cost-effective agentic applications and coding tasks. This model demonstrates superior performance across various benchmarks, offering significant advancements over its predecessors. Google also introduced Managed Agents and the Antigravity platform to streamline AI agent development and deployment.
Google has unveiled Gemini 3.5 Flash, the latest and fastest model in the Gemini 3.5 series, engineered for intelligent agents and coding. This new iteration offers significant improvements in speed and cost-efficiency, outperforming the previous Gemini 3.1 Pro across challenging benchmarks. It marks a "major leap" for AI agents, combining advanced intelligence with robust action capabilities.
The model demonstrates impressive performance with a 76.2% score on Terminal-Bench 2.1 for coding, 1656 Elo on GDPval-AA for agentic task performance, and high scores on MCP Atlas and CharXiv Reasoning for tool-use reliability and multimodal understanding, respectively. Notably, Gemini 3.5 Flash is four times faster for output tokens and can complete tasks at less than half the cost, with competitive pricing for input and output tokens. It features a substantial context window of over 1 million input tokens.
To facilitate development, Google introduced Managed Agents within the Gemini API, allowing developers to deploy full agents with a single API call. These agents can reason, use tools, and execute code within isolated Linux containers, maintaining state across sessions. This abstracts away the complexities of managing agent environments, a task that was previously manual.
Accompanying these advancements is Google Antigravity, an agent-first development platform. Antigravity 2.0, a standalone desktop application, enables the orchestration of multiple parallel agents and scheduled background automation. For developers, the Antigravity CLI offers instant agent creation without a GUI, while the SDK provides programmatic access for custom agent behaviors. These tools aim to streamline the process from idea to production-ready AI applications.
Enterprise partners are already leveraging Gemini 3.5 Flash. Shopify uses subagents for data analysis to enhance merchant growth forecasts, and Macquarie Bank is piloting it for complex customer onboarding. Salesforce is integrating it into Agentforce for automated enterprise tasks, while Ramp utilizes it for smarter OCR on invoices. Xero and Databricks are also deploying agents for sophisticated workflows, from gathering supplier data to real-time data monitoring and issue diagnosis.
Related articles
Build real agentic apps using CUGA: two dozen working examples on a lightweight harness
CUGA, IBM's open-source Agent Harness, simplifies building agentic applications by handling infrastructure, allowing developers to focus on tools and prompts. It offers pre-assembled components for planning, execution, and state management, significantly reducing development time. CUGA has topped agent benchmarks like AppWorld and WebArena.
OpenAI launches new initiative to help find and patch open source bugs
OpenAI has launched "Patch the Planet," a new initiative in partnership with cybersecurity firm Trail of Bits, to enhance the security of open-source projects. This program aims to assist maintainers in identifying and patching bugs, utilizing OpenAI's AI-powered security tools while reducing the burden on project teams.
PP-OCRv6 on Hugging Face: 50-Language OCR from 1.5M to 34.5M Parameters
Baidu has released PP-OCRv6, an advanced optical character recognition (OCR) model supporting 50 languages. Available on Hugging Face, this version significantly improves accuracy and efficiency across various parameter sizes, from 1.5 million to 34.5 million, marking a substantial leap in multilingual OCR technology.