Performance and Technical Specifications

Gemini 3.5 Flash is Google's latest model tier, designed to balance frontier-level intelligence with high-speed execution. It significantly outperforms the previous Gemini 3.1 Pro tier across key benchmarks, including 76.2% on Terminal-Bench 2.1 (coding) and 83.6% on MCP Atlas (tool-use reliability). The model features a 1,048,576-token context window and a 65,536-token maximum output.

Efficiency is a primary focus: the model is 4x faster on output tokens and introduces a "dynamic thinking" feature that automatically allocates additional compute resources for complex problems. Pricing is set at $1.50 per million input tokens and $9.00 per million output tokens, with cached input available at $0.15 per million.

Managed Agent Infrastructure

Google has moved beyond simple model inference by introducing "Managed Agents" within the Gemini API. This infrastructure abstracts the complexity of state management, tool calling, and iteration. Agents run within isolated Linux containers where files and state persist across multi-turn sessions, enabling long-horizon tasks that previously required manual orchestration.

This is supported by the "Antigravity" ecosystem, which includes:

  • Antigravity 2.0: A standalone desktop application for orchestrating parallel agents and background automation.
  • Antigravity CLI: A terminal-based interface for rapid agent creation.
  • Antigravity SDK: A programmatic harness for defining custom agent behaviors and hosting them on preferred infrastructure.

Enterprise Adoption

Early enterprise deployments demonstrate the model's utility in complex, multi-step workflows. Notable use cases include:

  • Shopify: Running parallel subagents for global merchant growth forecasting.
  • Macquarie Bank: Automating customer onboarding by reasoning over 100+ page documents.
  • Salesforce: Powering "Agentforce" to manage enterprise tasks with persistent context across tool calls.
  • Databricks: Utilizing agentic workflows to monitor real-time data, diagnose technical issues, and propose automated fixes.