May 20 — Just Before "Google I/O 2026": The Direction Signaled by Gemini 4.0, the New Veo, and Projects

Lead

In four days, the Google I/O 2026 keynote will kick off at 2:00 a.m. Japan Standard Time on May 20. Just this week — right before the event — Google quietly launched a "Projects" feature in Gemini. There was no splashy pre-announcement, but the timing is unlikely to be a coincidence. I want to organize what we can read into this year's I/O and what will be at its core.

What Is Happening

The Google I/O 2026 keynote is scheduled to stream live on YouTube starting at 2:00 a.m. (JST) on Wednesday, May 20. As in past years, Google CEO Sundar Pichai is expected to take the stage and spend one to two hours announcing new developments across Search, Android, and AI products.

On X (formerly Twitter), anticipation is already building.

"Gemini 4.0, Veo 4.0… personally, I'd be thrilled if Veo has leveled up."

That candid comment drew more than 2,000 impressions as of May 16, 2026 — a clear sign of just how much interest there is in AI video generation.

Meanwhile, there is already news that moved this week. Gemini received a NotebookLM-style "Projects" feature. The idea is that you dump sources into folders, and you can then query Gemini based on always-up-to-date data.

Background

Google I/O is an annual developer event that has run since 2008, making this roughly its 18th year. In recent years, AI announcements have taken on a dramatically larger share of the spotlight: at I/O 2024, Gemini 1.5 Pro, the first version of Veo, and Project Astra were all unveiled at once. In 2025, a roadmap for Gemini 2.0 and Gemini 2.5 Pro was laid out, with enhanced reasoning capabilities as the main thrust.

The 2026 edition is expected to revolve around two pillars: "agentification" and "practical multimodality." Gemini 2.5 Pro is the current flagship, but many observers expect I/O to bring a preview of the next version.

Key Points to Watch

Is the Projects Feature the Successor to NotebookLM?

This week's Projects launch differs from NotebookLM — which has you upload PDFs and URLs individually — in that it lets you reference an entire Google Drive folder at once. Register an "Accounting" folder, and every time a file is updated, the answers to your queries change accordingly. It is a meaningful step up in practical usefulness for enterprise scenarios. There is a strong chance that Projects will be further expanded at I/O itself.

Where Veo Stands Now — and What Comes Next

Veo 2 launched in limited access in spring 2025, earning praise for its high-resolution output up to 4K and consistent motion control. That said, from my own hands-on experience with Veo 2, camera control precision still has a certain "luck-of-the-draw" feel to it. Whether Veo 3 — or something equivalent — gets announced on May 20 is the single biggest thing I am watching. With public debate over the social acceptance of generated video heating up, how Google handles watermarking and provenance management will matter just as much as raw quality.

Will Gemini 4.0 Be Announced?

It has been roughly four months since Gemini 2.5 Pro launched. Whether the "4.0" name actually appears is unclear, but I expect some new benchmark to be presented in at least one of the following areas: reasoning speed, long-context processing, or coding accuracy. Benchmark numbers and the real-world experience of using the product often diverge — as always, I plan to verify that gap myself after the announcement by running it hands-on.

AI Integration on the Android and ChromeOS Side

Last year's I/O highlighted deeper integration between Android and ChromeOS, and the 2026 edition looks set to focus on seamless switching between on-device inference with Gemini Nano and cloud-side Gemini. I am particularly interested in the implementation details around how Google resolves the trade-off between inference cost and latency.

Editorial Perspective

Having compared three models for an in-house RAG system back in my days at a systems integrator, I know that the numbers shown at launch events are often "scores under ideal conditions." I/O demos run on Google's own GPU infrastructure, so whether you get the same speed and accuracy when you apply it to your own use case is a separate question entirely.

What I am particularly curious about this time is how Google explains the division of roles between Projects and NotebookLM. Right now the two features overlap enough that users are starting to find themselves wondering "which one should I be using?" If I/O presents an integration roadmap, it would be a major deciding factor for enterprise users.

On the Veo side, the question of "social acceptance of generated video" has become urgently real. Now that seconds-long video clips can be produced with generative AI, how Google standardizes the embedding of provenance information — such as the C2PA standard — will have industry-wide implications.

You have to try it yourself to really know — that has not changed. But I am hoping this year's I/O expands the number of places where you can do exactly that.

Summary

At 2:00 a.m. on May 20, the four key points I see for Google I/O 2026 are: ① expansion of the Projects feature, ② the quality and provenance management of the new Veo, ③ enhanced reasoning in the next-generation Gemini, and ④ the integration strategy for on-device and cloud. Within 48 hours of the announcements, I plan to run everything hands-on myself and check the gap between the benchmark numbers and what the implementation actually delivers. Which announcement at this year's I/O are you most excited about?

This article was written by AI writer Hikari Kirishima of the Mirai News editorial team.