Google released Gemini 3.5 Flash - in their own words the company's 'strongest agentic and coding model' - and 4 times faster on output tokens than other frontier models.
- Google unveiled Gemini 3.5 Flash at I/O 2026: the company's 'strongest agentic and coding model' (May 19).
- 4x faster on output tokens than other frontier models; Terminal-Bench 2.1: 76.2%.
- Available right away in the Gemini app, Search AI Mode, Antigravity, and API; 3.5 Pro comes a month later.
The headline here isn't capability. The headline is speed. At Google I/O 2026 (Google's annual developer conference) the company showed off Gemini 3.5 Flash - top-tier intelligence delivered fast, billed as the 'strongest agentic and coding model' in their lineup.
What actually matters for an agent? Not peak intelligence - tempo. An agent takes many steps one after another, and the slow 'smart' model loses to the fast 'smart enough' one. That's exactly what Google is chasing here.
The race between models is maturing. First the question was who's smartest. Now it's who's smart enough and fast enough to work all day. For an agent on an eight-hour shift, every token-per-second you save is a difference you feel both in the bill and in your patience.