Google's Gemini 3.5 Flash launched today with 300 tokens per second output speed while maintaining frontier-level intelligence comparable to larger, more expensive models.

The new model costs $1.50 per million input tokens and $9 per million output tokens, undercutting the previous 3.1 Pro model's $2 and $12 pricing. Google estimates companies using the most AI tokens could save a billion dollars annually by switching to the more efficient Flash variant.

Gemini 3.5 Flash substantially outperforms older Flash models on coding benchmarks including Terminal Bench and SWE-Bench Pro, while showing measurable improvements over Gemini 3.1 Pro. The model scores similarly to OpenAI's much larger GPT 5.5 on several evaluations.

"With post-training, we're really starting to unlock some of the value of the feedback we're getting from users," said Tulsee Doshi, senior director of product management for Gemini, citing improvements from Google's Antigravity IDE usage data.

The model excels at agentic workflows that require extended reasoning and tool use. On OSWorld-Verified, which tests how models handle general computing tasks, Gemini 3.5 Flash outperforms older Flash models and matches GPT 5.5 performance.

Gemini Spark brings persistent AI agents

Google also unveiled Gemini Spark, a dedicated AI agent that runs continuously in Google's cloud using the new 3.5 Flash model. Unlike chatbots tied to specific browser tabs, Spark operates across users' entire Google ecosystem.

Spark can monitor Gmail for specific emails, generate meeting summaries, and create daily digests by pulling context from Drive files and other Google services. The agent asks for approval before taking "high-stakes actions" and can send notifications or follow-up questions.

Doshi described using Spark internally to compile evaluation data for slide decks and track developmental milestones for her child. "It turned out beautifully," she said of the slide deck creation. "Probably better and in much less time than I would have been able to do."

Gemini 3.5 Flash is rolling out across the Gemini app, API, AI Studio, Android Studio, and Google's enterprise products. The Pro variant enters internal testing next month with release planned for June.