Particle.news

Google Ships Gemini 3.5 Flash as Default for Long‑Context Agent Workflows

It introduces a 1 million‑token context window, thought preservation and live search tools, signaling a push into Workspace and developer automation.

Overview

  • Google announced Thursday that Gemini 3.5 Flash is generally available and positioned as the new default Flash‑tier model for production use.
  • The model adds a 1 million‑token context window, a Mixture‑of‑Experts design and a thought preservation feature that keeps intermediate reasoning across multi‑turn sessions.
  • Google exposed Search AI Mode inside Gemini so the model can run autonomous live web searches and synthesize results, a capability the company says reached 97% factual accuracy in internal tests.
  • The launch includes pricing and platform access for developers with published token rates, API and AI Studio support, and three new coding agents for IDE completions, repo reviews and an Android Studio agent.
  • Project Astra will move to Workspace in Q3 2026 as a persistent, screen‑aware assistant and the release raises privacy, factuality and lock‑in questions developers and enterprises must evaluate.