Google Launched Gemini 3.5 Flash at I/O 2026 With Frontier Agentic Performance
Google has released Gemini 3.5 Flash at Google I/O 2026, the first model in its new Gemini 3.5 series. The model is available today to billions of users globally and targets long-horizon agentic tasks at a speed Google says is four times faster than other frontier models.
What Gemini 3.5 Flash Is
Gemini 3.5 Flash is Google's strongest agentic and coding model to date. Google positions it as a model that delivers flagship-level intelligence at Flash-series speed, without the latency tradeoff that has typically accompanied frontier performance.
According to Google's official announcement, 3.5 Flash outperforms Gemini 3.1 Pro on the following benchmarks:
- Terminal-Bench 2.1: 76.2%
- GDPval-AA: 1,656 Elo
- MCP Atlas: 83.6%
- CharXiv Reasoning (multimodal): 84.2%
On the Artificial Analysis Intelligence Index, 3.5 Flash lands in the top-right quadrant, meaning it scores at frontier intelligence levels while leading on output speed.
Where It Is Available
Gemini 3.5 Flash is live today across all major Google surfaces:
- Consumers: Gemini app and AI Mode in Google Search, globally
- Developers: Google Antigravity agent platform, Gemini API in Google AI Studio and Android Studio
- Enterprises: Gemini Enterprise Agent Platform and Gemini Enterprise
Google also confirmed Gemini 3.5 Pro is in development. It is already in internal use and is planned for public rollout next month.
The Antigravity Platform and Sub-Agent Architecture
The Gemini 3.5 launch is paired with Google Antigravity, its updated agent-first development platform. When running inside Antigravity, 3.5 Flash deploys collaborative sub-agents that execute parallel workstreams on complex tasks.
Google describes tasks that previously took developers days or auditors weeks as now completable in "a fraction of the time, often at less than half the cost of other frontier models."
Demonstrated use cases include:
- Renaming and categorizing unstructured assets automatically based on dynamic rules
- Synthesizing the AlphaZero paper and coding a fully playable game in six hours using two agents
- Migrating a legacy codebase to Next.js
- Building full branding concepts and interactive UX flows in parallel
Enterprise Partners Already Deploying It
Google disclosed six enterprise partners running Gemini 3.5 Flash in production or active pilots:
- Shopify: Running parallel sub-agents to analyze complex data for more accurate merchant growth forecasts at global scale
- Macquarie Bank: Piloting customer onboarding acceleration by reasoning over 100-plus page documents with low latency
- Salesforce: Integrating into Agentforce to automate enterprise tasks using multiple sub-agents with persistent context and complex multi-turn tool calling
- Ramp: Powering smarter OCR for complex invoice processing through multimodal understanding and historical pattern reasoning
- Xero: Deploying agents to autonomously manage multi-week workflows including supplier identification and 1099 tax form preparation
- Databricks: Using agentic workflows to monitor real-time data, diagnose issues, and propose solutions for data science teams
Gemini Spark: a Personal AI Agent Powered by 3.5 Flash
Google also announced Gemini Spark, a personal AI agent built on 3.5 Flash. It runs continuously, takes action on the user's behalf, and helps manage daily digital tasks under user direction.
Gemini Spark is entering limited rollout to trusted testers today. Google plans to bring a public beta to Google AI Ultra subscribers in the United States next week.
3.5 Flash is also now the default model powering the Gemini app and AI Mode in Search globally.
Safety Framework
Google states that Gemini 3.5 was developed under its Frontier Safety Framework. New safety measures include:
- Strengthened cyber and CBRN safeguards
- Reduced rate of harmful content generation
- Reduced rate of incorrect refusals on safe queries
- Interpretability tools that review the model's internal reasoning before it delivers a response
Developers evaluating Gemini 3.5 Flash alongside other frontier models can compare Gemini and competing AI models to assess performance across agentic and coding tasks.
