AI Agent Phones Arrive + The Gemini 3.5 Pro Guessing Game

Every leak points to Gemini 3.5 Pro going live today, July 17 — a 2M-token context window (double the field), input pricing that reportedly undercuts GPT-5.6, and a 'Deep Think' reasoning mode. The catch: Google hasn't confirmed a word. There's no model card, no pricing page, and nothing in the public API docs yet. 🤖 The date itself slipped from June after Google reportedly rebuilt the model following internal failures. If it ships today and delivers, it's a real shot at the top of the frontier. Until it's actually on the API, treat all of it as rumor, not fact. ⏳

The problem

When a launch date leaks from every direction except the company actually shipping the product, it's genuinely hard to tell rumor from confirmed fact.

What changed

Every leak points to Gemini 3.5 Pro going live July 17 with a 2-million-token context window, pricing that reportedly undercuts GPT-5.6, and a 'Deep Think' reasoning mode — following a slip from June after Google reportedly rebuilt the model after internal failures.

The catch

Google hasn't confirmed a word of it — there's no model card, no pricing page, and nothing in the public API docs yet, despite the wall-to-wall leak coverage.

What it means for you

Until it actually shows up in the API, treat every spec in this story as rumor, not fact — if it ships today and delivers, it's a real shot at the top of the frontier, but 'everyone says so except Google' is not the same as an official launch.

It's official: nubia unveiled the NaviX Ultra and is branding it the world's first AI agent smartphone, launching at WAIC 2026 in Shanghai today. 🚀 The difference versus every 'AI phone' so far is that this isn't a chatbot stapled onto the OS — the agent is baked into the system and can run multi-step tasks across apps on its own, powered by ByteDance's Doubao assistant. And it's real, not a concept: mass production is confirmed and it already picked up a 2026 SAIL award. 'World's first' is a marketing flex, but agentic-by-default is clearly where every phone is heading. The real test is reliability. 🤖

The problem

Every phone maker has slapped a chatbot onto their OS and called it an 'AI phone,' but none of them have actually let an AI agent run multi-step tasks across apps on its own.

What changed

nubia unveiled the NaviX Ultra at WAIC 2026 in Shanghai, branding it the world's first AI agent smartphone — the agent is baked into the system itself, powered by ByteDance's Doubao assistant, and mass production is already confirmed, backed by a 2026 SAIL award.

The catch

'World's first' is a marketing claim, and agentic-by-default is clearly where every phone maker is heading next — the real test isn't who says it first, it's whether the agent reliably completes tasks without supervision.

What it means for you

Don't judge this on the 'world's first' headline — watch the early reviews for reliability once it's in real hands, since that's the actual bar every agentic phone will need to clear.

Typed a wrong digit in the iPhone Calculator app? You don't need to hit C and start the whole sum over. 💡 Just swipe left or right across the number display — it deletes the last digit one at a time, like a backspace. It's been hiding in the default Calculator app for years, and it's one of those tiny things that quietly saves you a hundred restarts. ✅

The problem

Typing the wrong digit into the iPhone Calculator app has meant hitting 'C' and restarting the entire sum from scratch — a small but constant annoyance.

What changed

Swiping left or right across the number display in Calculator deletes the last digit one at a time, like a backspace — a feature that's been sitting in the default app for years.

The catch

It only works on the number display itself, so it won't help you edit an already-completed calculation — just the number you're currently typing.

What it means for you

Next time you fat-finger a digit mid-calculation, swipe instead of hitting clear — it's a tiny habit that saves a hundred restarts over time.

Everyone asks 'which AI is best?' — wrong question. Best for what? Here's the honest split. 🤖 ChatGPT is the default all-rounder with the best voice mode and biggest ecosystem. Gemini is unbeatable if you live in Google apps and Search, and its context window is massive. Claude is the pick for writing and coding — clean, careful output. ChatGPT still leads on usage at roughly 54% share, but 'most popular' isn't the same as 'best for your task.' Use the right tool for the job, not the one you're loyal to. 🧠

The problem

'Which AI is best' keeps getting asked as if there's one universal answer, when usage share and actual task performance often point in completely different directions.

What changed

The honest split: ChatGPT is the default all-rounder with the best voice mode and biggest ecosystem, Gemini is unbeatable if you live in Google apps and Search with its massive context window, and Claude is the pick for writing and coding with clean, careful output — even though ChatGPT still leads on usage at roughly 54% share.

The catch

'Most popular' and 'best for your task' are two different things — ChatGPT's usage lead reflects ecosystem and habit as much as it reflects which model actually does your specific job best.

What it means for you

Pick based on the task in front of you, not brand loyalty — Google-heavy workflows lean Gemini, writing and code lean Claude, and everything else defaults fine to ChatGPT.