What is Google I/O?
Google I/O is Google's annual event where the company presents its major new products for the year ahead. It is the equivalent of Apple's WWDC: two days during which Google's executives and engineers take to the stage to announce major strategic directions, new AI models, upcoming versions of Android, and everything relating to their products such as Chrome, Gmail, Search and YouTube.
This year, the event takes place on 19 and 20 May 2026 at Shoreline Amphitheatre in Mountain View, California. The opening keynote begins at 7pm French time. Everything is streamed live and free on io.google.
Why does this edition matter?
The context is unusual. In recent weeks, competition in artificial intelligence has hit hard on every front: new flagship models, breakthrough image generators, agentic tools, and even open-source models rivalling the closed giants. Claude Opus 4.7 and GPT-5.5 each demonstrated impressive capabilities in April 2026. Google, for its part, has stayed relatively quiet. Sundar Pichai (CEO of Alphabet, Google's parent company) is under maximum pressure to show that the Google ecosystem can hit back at the same level.
According to industry analysts, Google is said to have invested 185 billion dollars in AI in 2025, more than any other company worldwide. I/O 2026 is the opportunity to prove that this spending has produced something other than promises.
Google has confirmed three things for this edition: announcements around Gemini (their family of AI models), developer tools with an emphasis on "agentic coding" (AI that codes autonomously, like Claude Code or Codex), and updates for Android, Chrome, Cloud and their other products. Everything else falls under leaks and rumours.
Gemini 4: the likely big announcement
The most anticipated scenario is the unveiling of Gemini 4, the successor to the Gemini 3 family. Several credible leaks mention a score of 84.6% on ARC-AGI 2, a reasoning test known for resisting AI. By way of comparison, the most powerful current models plateau at around 50 to 60% on this benchmark.
If that figure is confirmed, Gemini 4 would on paper become the most advanced model in the world at complex reasoning, ahead of Claude Opus 4.7 and GPT-5.5. Note, however: figures leaked before an event of this kind are often inflated by people close to Google to generate anticipation. Independent benchmarks will be needed to validate them.
Veo 4: Google's video riposte
Google is also expected to present Veo 4, the new version of its AI video generator. Veo 3.1, the current version, already generates 8-second clips in 1080p with synchronised audio. Rumours suggest Veo 4 will be capable of generating 30-second sequences in 4K, with consistent characters and settings between shots, something still impossible today on most competing models.
>>For content creators and production studios, this would be a generational leap. The 30-second mark represents the minimum threshold for producing complete ads or narrative sequences without manual cutting.
Nano Banana Pro 2: the answer to Images 2.0
OpenAI struck hard with Images 2.0 on 21 April, crushing Google's Nano Banana Pro on generated text accuracy and speed. Unsurprisingly, Google is expected to unveil Nano Banana Pro 2 at I/O, their direct answer. Early leaks suggest the model would aim specifically to reclaim the Elo score lost on LM Arena and to fix the flaws in text rendering.
This is precisely the riposte we discussed in our recent article on the release of GPT-5.5.
Beyond the models: a unified OS and connected glasses
Google will not stop at AI models. Several other major announcements are expected:
- Aluminium OS: a new operating system for computers, intended to merge Android and ChromeOS into a single platform. If confirmed, it would be the biggest overhaul of Google's OS strategy in 10 years.
- Android XR and connected glasses: Google has entered a partnership with Warby Parker (an American glasses manufacturer) to produce smart glasses integrating Gemini. The model would weigh less than 50 grams and would be designed to compete with the Ray-Ban Meta. Samsung is also said to be working on a version of its own.
- Personal Intelligence: a new feature that connects Gemini to your Gmail, Google Photos, Drive, Calendar and YouTube history to create an assistant that knows your personal data. Announced for 2 billion users in more than 200 countries.
- Boston Dynamics and Atlas: the partnership unveiled in January 2026 between Google DeepMind and Boston Dynamics integrates Gemini into the humanoid robot Atlas. An on-stage demo could take place.
Several sources indicate that Google and Apple have signed a multi-year agreement: the next generation of Apple's AI models will be based on Gemini. Apple would use Gemini as a basis for training its own smaller models that run locally on iPhone and Mac. If this is confirmed at I/O, it is an earthquake in the balance of power with OpenAI, which had until now been Apple's main partner.
What is likely, what is not
To help you sort out what will actually be announced from what will stay in the corridors, here is an honest ranking based on the reliability of the sources:
| Expected announcement | Likelihood |
|---|---|
| Gemini 4 (general announcement) | Almost certain |
| Veo 4 (video generation) | Very likely |
| Nano Banana Pro 2 (image generation) | Very likely |
| Android 17 (final updates) | Certain |
| Aluminium OS (new operating system) | Likely |
| Android XR glasses with Warby Parker | Likely |
| Atlas demo (Boston Dynamics) | Possible |
| Pixel 11 (new phone) | Unlikely |