Google I/O 2026: Gemini 4.0, Omni, Android XR and Aluminium OS — everything Google just announced
🔎 Why Google I/O 2026 is the most important event since the launch of Android
On May 19 and 20, 2026, Google held an I/O that feels like a tipping point. Not an incremental keynote, but a systemic overhaul: a new flagship model, a novel generative video model, XR glasses, a hybrid OS, and a persistent AI agent.
The context is tense. OpenAI dominates the agentic leaderboard with GPT-5.5 at 98.2, and Anthropic pushes Claude Opus 4.7 (Adaptive) to 94.3. Google had to respond on all fronts — model, agent, device, ecosystem. That's exactly what it did, spectacularly so according to feedback from Wired and Android Central.
The question is no longer whether AI will be integrated everywhere, but how Google plans to prevent OpenAI and Anthropic from dictating the terms of this integration.
The essentials
- Gemini 4.0: new flagship model with 2 million context tokens, positioned against GPT-5.5 and Claude Opus 4.7.
- Gemini Omni: generative video model leaked ahead of the keynote, capable of generating and understanding native video content.
- Android XR Glasses: mixed reality glasses developed with Samsung and XREAL, Google's first XR hardware since Glass.
- Aluminium OS: merger of ChromeOS and Android into a single operating system, ending the historic separation.
- Gemini Spark: persistent AI agent that lives in the background of Android 17, Chrome, and Google Cloud.
- Googlebooks: integration of Gemini into the book ecosystem, Android becomes an "intelligence system".
Recommended tools
| Tool | Main usage | Price (June 2025, check on google.com) | Ideal for |
|---|---|---|---|
| Gemini 4.0 | Flagship LLM model | Free (limited) / Advanced starting at 21.99€/month | Research, long analysis, development |
| Gemini Omni | Generative video | Integrated into Gemini Advanced | Creators, editing, visual prototyping |
| Android XR Glasses | Mixed reality | Unannounced price (estimated 800-1200€) | Mobile productivity, hands-free navigation |
| Gemini Spark | Persistent agent | Included in the Android 17 ecosystem | Daily automation, agentic tasks |
Gemini 4.0: the model that must catch up to GPT-5.5
Gemini 4.0 is the heart of this keynote. Google positions it as its most capable model, with a context window of 2 million tokens — a significant leap from the previous generation.
The numbers that count
According to the internal benchmarks presented by Google during the keynote, reported by Aixploria, Gemini 4.0 ranks at the top of the class. It doesn't beat GPT-5.5 (98.2 in agentic), but it closes part of the gap.
The 2M token context window is the major selling point. This makes it possible to ingest entire codebases, complete books, or dozens of hours of audio transcriptions. A concrete advantage that neither GPT-5.5 nor Claude Opus 4.7 offer at this scale in their public versions.
Competitive positioning
In the Gemini vs ChatGPT vs Claude comparison, the central question becomes: is long context enough to compensate for a potentially lower agentic score? For software development, the answer is nuanced. Gemini 4.0 excels at analyzing large codebases, but GPT-5.5 remains ahead on complex autonomous agent tasks.
For users looking for the best free LLMs, the free version of Gemini 4.0 offers an unbeatable quality-to-context ratio. This is a strategic point for Google: the entry point is wide, and monetization comes through Advanced and Cloud.
Gemini Omni : generative video finally arrives at Google
Gemini Omni is probably the most surprising announcement — and the most poorly managed in terms of communication. The model leaked before the keynote, according to Pasquale Pillitteri, forcing Google to accelerate its presentation.
What Gemini Omni actually does
Omni is a native multimodal model for video. Not simply a model that analyzes existing videos, but one that generates new ones from text prompts or images. Google describes it as an "omnidirectional perception model", a term that echoes the work of the Omni-Captioner benchmark published on arXiv in 2025 for evaluating detailed multimodal perception.
The distinction is important: Omni is not Sora. It is natively integrated into the Gemini ecosystem, not isolated in a research lab. A user can ask Gemini to generate a video sequence, then analyze it, modify it, caption it — all within the same flow.
Impact on the creative market
For creators, Omni changes the game if it delivers on its quality promises. Video generation remains the domain where the technical bar is the highest. If Google delivers a consistent, fluid, and controllable model, this puts direct pressure on OpenAI, which has not yet rolled out Sora at scale.
Android XR Glasses: the glasses Google should have released 10 years ago
Google Glass was a commercial failure in 2014. Android XR Glasses is the redemption attempt, and this time, Google isn't going it alone.
The Samsung, XREAL, Google trio
According to Android Authority, Android XR Glasses are the result of a partnership with Samsung for the hardware and XREAL for the optics. The design evokes classic sunglasses, not a bulky VR headset.
The idea: a discreet AR screen that overlays information without cutting the user off from the real world. Notifications, navigation, real-time translation, Gemini summaries — all controlled by voice or minimal gestures.
Why it's different this time
Three factors have changed since Glass. AI is now capable of understanding context without the user formulating a specific query. Micro-displays have improved in brightness and resolution. And the market has been educated by Apple Vision Pro and Meta Ray-Ban, which have proven that there is a demand for lightweight AR.
The danger for Google: launching a niche product at a high price without a compelling app ecosystem. That is exactly what killed Glass.
Aluminium OS : the end of ChromeOS as we know it
This is the most structural announcement. Aluminium OS merges ChromeOS and Android into a single operating system. Not just simple coexistence, but integration at the kernel level.
Why now
ChromeOS was designed for an era where the browser was the center of everything. In 2026, Android apps, on-device AI, and rich multitasking make this separation artificial. A Chromebook cannot run all Android apps correctly. An Android tablet lacks the productivity of a native keyboard.
Aluminium OS solves this by unifying the two worlds. Android apps run natively. The Chrome browser remains central but is no longer the only paradigm. The interface adapts to the form factor — tablet, laptop, touchscreen.
The risks of this merger
Google has a bad habit of killing products. The question on developers' minds, according to reactions reported by Android Central, is simple: will current ChromeOS apps be compatible? And will the lightweight nature of ChromeOS (booting in a few seconds, seamless updates) survive the merger?
If Google manages the transition correctly, Aluminium OS could become the most versatile OS on the market. If not, it will be another Fuchsia — promising on paper, invisible in reality.
Gemini Spark : the agent that refuses to disappear
Gemini Spark is perhaps the most strategic long-term announcement. It is a persistent AI agent that lives in the background on your device.
How Spark works
Unlike a classic chatbot that you open, query, and close, Spark remains active. It observes your habits, anticipates your needs, and acts proactively. Concrete examples given during the keynote: automatically summarizing your Google Meet meetings, preparing email replies based on your history, suggesting schedule changes based on your calendar.
This is exactly the "agentic" paradigm that the LLM Agentic benchmark measures. GPT-5.5 dominates with 98.2, but Google is betting that system integration is worth more than the raw score. A less powerful but always-available, always-contextualized agent could be more useful than a supremely powerful agent locked in a browser tab.
The question of trust and privacy
A persistent agent that observes everything is a privacy nightmare by default. Google insisted on on-device processing for sensitive tasks and end-to-end encryption. But Google's history with data makes this promise hard to swallow for savvy users.
If you are sensitive to these issues, Anthropic's Claude Opus 4.7 or self-hosted DeepSeek V4 Pro models offer more granular control over your data.
Googlebooks and Android 17: the intelligence system
The integration of Gemini into Android 17 goes beyond simple "smart suggestions." Google announced Googlebooks, an ecosystem where AI becomes the reading, analysis, and synthesis layer for all textual content.
Android as an intelligence system
As detailed in our article on Google Gemini Intelligence et Googlebooks : Android devient un « système d'intelligence », the idea is to transform Android from an OS that runs apps into an OS that understands and acts on your data. Gemini is no longer an assistant you consult — it is the connective tissue of the interface.
In practice: you highlight a passage in a Google Play Books book, Gemini automatically connects it to your notes, your emails, your Google Workspace documents. The context is no longer limited to an app; it spans the entire system.
Chrome and Cloud: same logic
Chrome integrates Gemini at the level of the search engine and the browser itself. Not an optional sidebar, but automatic page summaries, contextual translations, and predictive navigation. On the Cloud side, Gemini 4.0 models are deployed on Vertex AI with the 2M token context window, which opens up document analysis use cases that were impossible six months ago.
For developers who want to integrate these capabilities without going directly through Google Cloud, the APIs IA gratuites via OpenRouter or Groq remain relevant alternatives for prototyping.
Competitive impact: Google vs OpenAI vs Anthropic
This keynote should be read through the prism of the model war. The landscape in June 2025 is clear: GPT-5.5 dominates, Claude Opus 4.7 innovates with adaptive reasoning, and Google plays the system integration card.
Where Google wins
The 2M token context of Gemini 4.0 is a measurable advantage. The Android-Chrome-Cloud ecosystem gives Google a distribution surface that no competitor possesses. Gemini Omni in generative video, if it is good, fills a real gap in Google's offering. Aluminium OS and Android XR show that Google is thinking beyond the model — it is thinking platform.
Where Google loses
The agentic scores speak for themselves. GPT-5.5 at 98.2, Gemini 3 Pro Deep Think at 95.4, Claude Opus 4.7 at 94.3. Even if Gemini 4.0 improves these numbers, catching up with OpenAI on pure agentic reasoning is a different challenge from system integration. In the detailed comparison, Google's strength remains long context and the ecosystem, not pure reasoning.
Anthropic, for its part, retains the advantage of trust and safety. Claude is perceived as more reliable on critical tasks, less verbose, more precise. Gemini Spark, with its background persistence, could exacerbate this perception problem.
❌ Common mistakes
Mistake 1: Confusing Gemini Omni with a simple video editing tool
Omni is not CapCut. It is a generative model that creates video from text or images, with a native understanding of movement, time, and storytelling. Comparing it to an editing tool is like comparing DALL-E to Photoshop.
Mistake 2: Thinking Aluminium OS is coming immediately to all Chromebooks
The ChromeOS/Android merger is an 18- to 24-month migration process. Existing Chromebooks will receive an update, but the full experience (native Android apps, adaptive interface) will require compatible hardware. Don't throw away your current Chromebook, but don't expect a magical change tomorrow.
Mistake 3: Using Gemini Spark like a classic chatbot
Spark is designed to act in the background. If you open it like a chat and ask it one-off questions, you are missing out on 90% of its value. Set it up for recurring tasks (meeting summaries, email sorting, project monitoring) and let it work.
Mistake 4: Ignoring alternatives for development
Gemini 4.0 is excellent for analyzing large codebases thanks to its context. But for pure agentic coding, GPT-5.3 Codex (87 in agentic) or DeepSeek V4 Pro Max (88 in general) remain competitive. Check out our guide to the best LLMs for coding before locking yourself into a single ecosystem.
❓ Frequently Asked Questions
Does Gemini 4.0 replace Gemini 3.1 Pro?
No. Gemini 4.0 is the new flagship, but Gemini 3.1 Pro (score 92) remains available as a lighter and faster model, suitable for tasks that do not require 2M tokens of context. Google maintains a portfolio strategy, not a single replacement.
Is Gemini Omni available now?
Omni was announced in limited preview at I/O. Access is restricted to Gemini Advanced subscribers and Vertex AI developers. General availability is expected in the third quarter of 2026, according to the keynote's indications.
Will Aluminium OS kill ChromeOS?
Yes, eventually. Google confirmed that ChromeOS, as a distinct brand, will gradually be absorbed by Aluminium OS. But the transition will be gradual, and the Chrome user experience (browsing, extensions, PWA) will be preserved. Don't panic if you are on a Chromebook.
Does Gemini Spark respect privacy?
Partially. Google promises on-device processing for sensitive tasks and end-to-end encryption for Cloud queries. But a persistent agent, by definition, needs to access your data. If privacy is your absolute priority, self-hosted models like DeepSeek V4 Pro or Kimi K2.6 are more suitable.
Do the Android XR Glasses replace the smartphone?
Not in the foreseeable future. Google positions the glasses as a complement, not a substitute. They offload notifications and quick tasks from the phone, but the smartphone remains the main hub for complex tasks. It's a similar strategy to Meta's with the Ray-Ban.
✅ Conclusion
Google I/O 2026 is the keynote of a giant that refuses to let AI become a market dominated by startups. Gemini 4.0 with its 2M tokens, Omni in video, Spark as a persistent agent, Aluminium OS as system unification — the message is clear: AI doesn't live in a chatbot, it lives in the OS. It remains to be seen whether the execution will match the ambition. To follow the evolution of these models, check out our ChatGPT vs Gemini comparison updated in real time.