Google I/O 2026: small models now rival last year's flagship
What was announced?
At Google I/O 2026 on May 19, Google made roughly one hundred announcements. Two model releases stood out. Gemini 3.5 Flash is a flash-class model: small, fast, and cheap to serve. Google positions it as rivaling flagship-level intelligence. Gemini Omni is a new multimodal model built to take any kind of input.
One correction is worth making early. Some quick recaps described an on-device “Gemini Spark.” That is not accurate. Gemini Spark, also announced at I/O, is a cloud agent that runs on Google’s servers. It gets its own Signal. Nothing about Spark runs on your phone.
What does it mean for personal, on-device writing?
The headline is not any single model. It is the curve. A flash-class model now rivals what counted as flagship intelligence a year ago. This has happened almost every cycle: the frontier moves up, and the efficient tier absorbs last year’s frontier.
For personal writing, the efficient tier is the one that matters. Drafting an email, a follow-up, a short update for your manager — none of this needs frontier reasoning. It needs solid language ability plus knowledge of you. The language ability keeps shrinking in cost and size. That means it keeps moving toward the device in your pocket.
Gemini 3.5 Flash itself is still a cloud model. But the efficiency it demonstrates is the same force that improves Apple’s on-device foundation models, and every other small model, year over year. Intelligence per watt keeps climbing. Phones are where that climb pays out.
Where does ILURA stand?
This changes nothing for ILURA users today. ILURA does not use Gemini. It runs on Apple Intelligence, on the iPhone, and that did not move on May 19.
But the trendline is ILURA’s long-term argument. ILURA’s bet is that the base model becomes a commodity: it ships with your phone and improves with every OS release. What never ships with the phone is knowledge of you. That is the layer ILURA builds — role playbooks for how you write in each part of your life, trained by your corrections, stored on the device, with no account involved. The smarter the on-device base gets, the more that layer is worth. I/O 2026 says the base is getting smarter, fast.
Read the signal through ILURA
Platform news matters when it changes what users expect from personal AI. ILURA reads these shifts through one lens: private agents trained by the user on iPhone.
- What becomes possible?
- What should stay user-controlled?
- What belongs on device?
Try it now
Put this to work on a real message.
Open ILURA, bring in a message you actually need to handle today, and get it done in your voice — free, on device, no account. It learns the preference, so the behavior carries to the next one.
Free to start · No account · Data Not CollectedQuick answers
- Is Gemini 3.5 Flash an on-device model?
- No. Gemini 3.5 Flash is a cloud model served from Google's infrastructure. The on-device signal is indirect: flash-class models reaching flagship quality shows how quickly efficient AI is improving.
- Does Gemini 3.5 Flash change anything for ILURA users?
- Not directly. ILURA runs on Apple Intelligence, not Gemini. But the same efficiency curve is the reason on-device writing keeps improving with every iPhone and iOS release, at no extra cost.