🔥 This Week’s Highlights
Anthropic owned the week. Claude Sonnet 5 quietly became the default model for every Free and Pro user — meaning most people talking to Claude just got a big upgrade without changing a thing — and its powerhouse sibling Fable 5 came back online worldwide after a dramatic 19-day, government-ordered blackout. Everyone else shipped genuinely useful consumer stuff too: Copilot Vision went free for all, Grok let anyone build a talking voice agent in two minutes, Google taught Slides to build a whole deck from one sentence, and Gemini’s Spark agent landed on the Mac. Hardware was quieter — the big machines are still the ones from early June — but Amazon confirmed more on-device AI is coming to Alexa. Here’s what was worth your attention.
🤖 AI Models & Features
Claude Sonnet 5 Becomes the Default Model for Free & Pro Users
🧠 News & Announcements · High Priority · 5 min to try
This is the week’s biggest deal for regular users: Anthropic launched Claude Sonnet 5 on June 30 and made it the default for Free and Pro. You didn’t have to opt in — your everyday Claude chats now run on a model that closes most of the gap to the pricier Opus tier, especially on multi-step “agentic” tasks where Claude uses a browser or terminal on its own. It also hallucinates less and resists prompt-injection attacks better. Just open Claude and give it something genuinely multi-step to feel the difference.
Claude in Chrome Is Now Free for Everyone
🌐 Products & Platforms · High Priority · 15 min to try
The browser extension that lets Claude actually read and click through web pages graduated from paid preview to a full, free release on July 1. It’s one of the most hands-on ways to try an AI agent that operates your browser — comparing options, pulling info off sites, filling forms. Install it, sign in, and start with low-stakes tasks so you can watch what it does before trusting it with anything sensitive.
Claude Fable 5 Returns Worldwide After a 19-Day Suspension
🔓 News & Announcements · Medium Priority · 5 min to try
Anthropic’s most powerful model came back online globally on July 1 after US export controls — triggered by a jailbreak Amazon researchers found — pulled it offline for 19 days. It returned with a new safety classifier that blocks the specific exploit in 99%+ of cases, though some legitimate security-adjacent coding prompts now get rerouted to Opus. Even if you never used Fable 5, this is the clearest case yet of governments directly shaping which AI models you can access.
Claude Science: A Research Workspace Opens to All Paid Plans
🔬 Products & Platforms · Medium Priority · 30 min to try
Claude Science is a new research hub that wires Claude into dozens of scientific databases, now in beta for all paid plans. You don’t need to be a scientist to benefit — it’s a great way to get grounded, source-linked explanations of technical topics. Anthropic also opened a grant program handing out up to $30,000 in API credits to 50 research projects (applications close mid-July).
ChatGPT Gets a Sharper Speech-to-Text System on All Plans
🎙️ Products & Platforms · Medium Priority · 5 min to try
OpenAI rolled out a new speech-to-text engine to every ChatGPT plan, and it’s noticeably better at accents, non-English speech, and tricky names or technical terms. If you’ve avoided dictating because transcription kept mangling your words, give it another shot — tap the mic and talk. It’s a small upgrade that quietly makes ChatGPT nicer to use on the go.
ChatGPT’s Personal Finance Dashboard Comes to Android
💰 Products & Platforms · Low Priority · 15 min to try
ChatGPT’s bank-linked finance dashboard — ask it plain-language questions about your spending — reached Android for US Pro and Plus users after debuting on web and iPhone. It nudges ChatGPT toward being an everyday money assistant. As always with anything touching bank data, review exactly what you’re connecting and which permissions you grant.
ChatGPT Retires Pulse and Folds It Into Upgraded Scheduled Tasks
⏰ Products & Platforms · Medium Priority · 15 min to try
OpenAI retired Pulse (its fixed daily briefing) around July 1 and merged it into a more flexible Scheduled Tasks hub. Instead of one morning digest, you can now set up multiple recurring jobs — “every morning at 7am, summarize the top AI headlines,” say. It’s one of the easiest ways to make ChatGPT work for you without prompting it each time. Pro users keep Pulse for a two-week transition.
Gemini Spark Lands on Mac and Connects to Instacart, OpenTable & More
✨ Products & Platforms · High Priority · 30 min to try
Google’s always-on Gemini Spark agent took two real steps forward: it shipped as a native Mac app (beta, Ultra subscribers, US) that can act on your files and handle desktop chores, and it plugged into everyday services like Instacart, OpenTable, Canva, Dropbox, Google Tasks/Keep, and Zillow Rentals. It can even watch a topic — sports scores, weather — and ping you when something changes. The catch: it’s Ultra-tier, so not free.
Google Slides Builds a Whole Deck From a Single Prompt
📊 How-To Guide · Medium Priority · 15 min to try
Gemini in Google Slides now generates a complete, editable slideshow from one prompt — layout, text, and images — instead of one slide at a time. Gemini in Sheets got a companion trick: one-click repair of a broken formula. If you live in Google Workspace, these turn genuinely tedious chores into a single sentence. Open a blank Slides file and describe the deck you want.
Google’s Nano Banana 2 Lite: Images in Four Seconds for Pennies
🍌 Products & Platforms · Medium Priority · 15 min to try
Google’s new Nano Banana 2 Lite is built for speed and volume — about four seconds per image at roughly $0.034 per 1,000 images — via Google AI Studio and the Gemini API. It’s not the model for a single gorgeous hero image; it’s the one for cranking out lots of thumbnails, mockups, or drafts cheaply. Google also widened access to its Omni Flash video model at $0.10 per second.
Microsoft Copilot Vision Comes to All Plans
👁️ Products & Platforms · Medium Priority · 5 min to try
As of July 1, Copilot Vision is free on every plan: attach a photo or PDF and Copilot will read it and reason about what’s inside. It’s a genuinely practical everyday tool — summarize a contract, decode an error screenshot, make sense of a chart or receipt. Open Copilot, attach a file, and ask something specific.
Copilot Now Edits Your Word Docs Directly and Adds FLUX.2 Slide Images
📝 Products & Platforms · Medium Priority · 15 min to try
Two nice Office upgrades: Copilot can now edit your Word document directly instead of asking permission for each change, and PowerPoint gained a FLUX.2-powered image generator for much better AI visuals in slides. If you spend your day in Office, both shave off real friction. Try asking Copilot to “tighten this section” in Word, or generate a custom slide image instead of hunting for stock art.
Grok Voice Agent Builder Opens to Everyone — Build a Voice AI in Minutes
🗣️ Products & Platforms · High Priority · 30 min to try
xAI opened Grok’s Voice Agent Builder to everyone on July 1 — a no-code way to spin up a talking AI phone agent in about two minutes, with 80+ voices, voice cloning, and roughly $0.05/minute calls. Every account gets a free test phone number, which makes it one of the most approachable ways to actually experience a voice agent. Describe what you want it to do, pick a voice, and call it.
GPT-5.6 Family (Sol, Terra, Luna) Previewed — Broad Release Pending
🚀 News & Announcements · High Priority · 5 min to try
OpenAI’s next big model line comes in three tiers — Sol (top), Terra (mid), Luna (fast/cheap) — but it’s still limited to about 20 vetted partners while a US government pre-release review is finalized. Analysts expect broad access mid-to-late July, with Terra positioned as the value pick most p…