구글 I/O 2026: AI 에이전트 최적화 'Gemini 3.5 Flash' 공개

Editors Pick Agentic AI Technology AI Shorts Artificial Intelligence Applications Language Model New Releases Software Engineering Staff Tech News Google just released Gemini 3.5 Flash at Google I/O May, 2026. It is the first Gemini 3.5 model. The series combines frontier intelligence with action. Google calls it a major leap for intelligent agents. The Flash tier has historically been faster and cheaper. 3.5 Flash outperforms Gemini 3.1 Pro on challenging benchmarks. The previous premium tier has now been surpassed. What the Benchmarks Say Gemini 3.5 Flash scores 76.2% on Terminal-Bench 2.1. That benchmark tests coding performance. It scores 1656 Elo on GDPval-AA. That measures real-world agentic task performance. It scores 83.6% on MCP Atlas. MCP Atlas measures scaled tool-use reliability. It scores 84.2% on CharXiv Reasoning. That benchmark tests multimodal understanding. Gemini 3.5 Flash is 4x faster on output tokens. Tasks often complete at less than half the cost. Official pricing is $1.50 per million input tokens. Output tokens cost $9.00 per million. Cached input is priced at $0.15 per million. The context window is 1,048,576 input tokens. Maximum output is 65,536 tokens. Supported inputs are text, image, audio, and video. The knowledge cutoff is January 2026. Dynamic thinking is on by default. The model auto-allocates more compute for harder problems. Built for Agentic and Long-Horizon Tasks Here ‘Agentic' means the model plans, calls tools, and iterates. It completes multi-step goals, not single questions. ‘Long-horizon' means that loop runs for extended periods. Google introduced Managed Agents in the Gemini API. One API call spins up a full agent. It reasons, uses tools, and executes code. The environment runs inside an isolated Linux container. Files and state persist across follow-up calls. This enables seamless multi-turn agent sessions. Previously, managing agent state and environments was manual. The Managed Agents API abstracts that infrastructure entirely. The Antigravity Ecosystem Google Antigravity is its agent-first development platform. It takes ideas to production-ready apps. Antigravity 2.0 is a new standalone desktop app. It orchestrates multiple agents running in parallel. Dynamic subagents handle parallelized workflows. Scheduled tasks enable background automation. Integrations cover Google AI Studio, Android, and Firebase. The Antigravity CLI is for terminal-based developers. It creates agents instantly, without a GUI. Google encourages Gemini CLI users to migrate now. The Antigravity SDK gives programmatic access to the harness. You can define custom agent behaviors with it. Host agents on infrastructure of your choice. Real-World Enterprise Deployments According to Google, several enterprise partners are already running 3.5 Flash. Shopify runs subagents in parallel for data analysis. It powers more accurate merchant growth forecasts globally. Macquarie Bank is piloting it for customer onboarding. The model reasons over complex 100+ page documents. It retrieves information and makes reliable recommendations. Salesforce is integrating 3.5 Flash into Agentforce. It automates enterprise tasks using multiple subagents. Subagents retain context across complex, multi-turn tool calling. Ramp uses it for smarter OCR on invoices. It pairs multimodal understanding with historical pattern reasoning. Xero deploys agents for complex, multi-week workflows. One example is gathering supplier data for 1099 forms. Databricks uses agentic workflows for real-time data monitoring. The model diagnoses issues and proposes fixes for engineers. Check out the Technical details . Also, feel free to follow us on Twitter and don’t forget to join our 150k+ ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us Michal Sutter + posts Bio Michal Sutter is a data science professional with a Master of Science in Data Science from the University of Padova. With a solid foundation in statistical analysis, machine learning, and data engineering, Michal excels at transforming complex datasets into actionable insights. Michal Sutter Upstash for Redis vs Supabase vs Neon: Which One Fits Vibe Coding Workflows in 2026? Michal Sutter Google Launches Antigravity 2.0 at I/O 2026: A Standalone Agent-First Platform with CLI, SDK, Managed Execution, and Enterprise Support Michal Sutter Vercel Labs Introduces Zero, a Systems Programming Language Designed So AI Agents Can Read, Repair, and Ship Native Programs Michal Sutter Enterprise AI Governance in 2026: Why the Tools Employees Use Are Ahead of the Policies That Cover Them Michal Sutter Google DeepMind Introduces an AI-Enabled Mouse Pointer Powered by Gemini That Captures Visual and Semantic Context Around the Cursor Michal Sutter OpenAI Introduces Daybreak: A Cybersecurity Initiative That Puts Codex Security at the Center of Vulnerability Detection and Patch Validation Michal Sutter Best Vector Databases in 2026: Pricing, Scale Limits, and Architecture Tradeoffs Across Nine Leading Systems Michal Sutter OpenClaw vs Hermes Agent: Why Nous Research's Self-Improving Agent Now Leads OpenRouter's Global Rankings Michal Sutter NVIDIA AI Just Released cuda-oxide: An Experimental Rust-to-CUDA Compiler Backend that Compiles SIMT GPU Kernels Directly to PTX Michal Sutter OpenAI Introduces MRC (Multipath Reliable Connection): A New Open Networking Protocol for Large-Scale AI Supercomputer Training Clusters Michal Sutter Google Adds Event-Driven Webhooks to the Gemini API, Eliminating the Need for Polling in Long-Running AI Jobs Michal Sutter Microsoft Research's World-R1 Uses Flow-GRPO and 3D-Aware Rewards to Inject Geometric Consistency Into Wan 2.1 Without Architectural Changes Michal Sutter Cursor Introduces a TypeScript SDK for Building Programmatic Coding Agents With Sandboxed Cloud VMs, Subagents, Hooks, and Token-Based Pricing Michal Sutter Top 10 KV Cache Compression Techniques for LLM Inference: Reducing Memory Overhead Across Eviction, Quantization, and Low-Rank Methods Michal Sutter smol-audio: A Colab-Friendly Notebook Collection for Fine-Tuning Whisper, Parakeet, Voxtral, Granite Speech, and Audio Flamingo 3 Michal Sutter xAI Launches grok-voice-think-fast-1.0: Topping τ-voice Bench at 67.3%, Outperforming Gemini, GPT Realtime, and More Michal Sutter Google DeepMind Introduces Vision Banana: An Instruction-Tuned Image Generator That Beats SAM 3 on Segmentation and Depth Anything V3 on Metric Depth Estimation Michal Sutter OpenAI Releases GPT-5.5, a Fully Retrained Agentic Model That Scores 82.7% on Terminal-Bench 2.0 and 84.9% on GDPval Michal Sutter Next Leap to Harness Engineering: JiuwenClaw Pioneers ‘Coordination Engineering' Michal Sutter OpenAI Scales Trusted Access for Cyber Defense With GPT-5.4-Cyber: a Fine-Tuned Model Built for Verified Security Defenders Michal Sutter xAI Launches Standalone Grok Speech-to-Text and Text-to-Speech APIs, Targeting Enterprise Voice Developers Michal Sutter A Coding Tutorial for Running PrismML Bonsai 1-Bit LLM on CUDA with GGUF, Benchmarking, Chat, JSON, and RAG Michal Sutter Top 19 AI Red Teaming Tools (2026): Secure Your ML Models Michal Sutter A Coding Guide to Build a Production-Grade Background Task Processing System Using Huey with SQLite, Scheduling, Retries, Pipelines, and Concurrency Control Michal Sutter Google AI Launches Gemini 3.1 Flash TTS: A New Benchmark in Expressive and Controllable AI Voice Michal Sutter A Coding Implementation of Crawl4AI for Web Crawling, Markdown Generation, JavaScript Execution, and LLM-Based Structured Extraction Michal Sutter Google AI Research Proposes Vantage: An LLM-Based Protocol for Measuring Collaboration, Creativity, and Critical Thinking Michal Sutter Meta AI and KAUST Researchers Propose Neural Computers That Fold Compu