Descriptions:
Matt Wolfe attended Google IO 2026 in person and delivers a firsthand breakdown of the week’s most significant announcements, narrowing Google’s own “100 things we announced” list down to the highlights most likely to affect developers, creators, and everyday AI users.
On the model front, Google unveiled the Gemini 3.5 family. The immediately available Gemini 3.5 Flash is positioned as a speed-and-cost leader: priced at $150 per million input tokens and $9 per million output tokens via API — compared to Claude Opus 4.7 at $5/$25 and GPT-5.5 at $5/$30 — while benchmarking on par with those models on agentic tasks and running more than twice as fast as Gemini 3.1 Pro and over three times faster than GPT-5.5. Gemini 3.5 Pro is still forthcoming. The second new model, Gemini Omni, is described by the team Wolfe interviewed as a “nano banana for video” — a multimodal generation model that accepts video, image, audio, and text inputs today, with plans to support any-to-any output in the future, grounded in world knowledge similarly to how image generation models produce research-backed infographics.
The most contested announcement is Google’s expansion of AI Overviews in Search, including interactive visual responses and a new “Ask YouTube” feature that surfaces answers without requiring users to click into videos or websites. Wolfe walks through the structural tension this creates: AI Overviews are built on content from publishers and creators, yet they reduce the traffic incentive for those same creators to keep producing content — raising questions about whether Google can sustain the model long-term.
📺 Source: Matt Wolfe · Published May 22, 2026
🏷️ Format: News Analysis
