• Everyday AI
  • Posts
  • Ep 821: Claude Desktop Gets Upgrade, New Open Source Model Shocks, ChatGPT Desktop Gets Better and 7 More AI Features You Can Use Today

Ep 821: Claude Desktop Gets Upgrade, New Open Source Model Shocks, ChatGPT Desktop Gets Better and 7 More AI Features You Can Use Today

Microsoft CEO Satya Nadella called Anthropic's Fable "editorially controlled," Kimi K3 became the world's No. 3 AI model, and the EU ordered Google to open Android and Search. And more.

Author

Everyday AI
July 17, 2026

 

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: Claude Desktop, ChatGPT Work, and open-source AI all got major upgrades this week. Here are the features you should actually be using. Give today’s show a watch/read/listen.

🕵️‍♂️ Fresh Finds: Google Search AI added app integrations, NVIDIA launched Nemotron 3 Embed, and OpenAI brought PR reviews to Codex. And more. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: Microsoft CEO Satya Nadella called Anthropic's Fable "editorially controlled," Kimi K3 became the world's No. 3 AI model, and the EU ordered Google to open Android and Search. And more. Read on for Byte Sized News.

💪 Leverage AI: Kimi K3 just reshaped the AI model race, while Claude Desktop and ChatGPT Work both got major upgrades. Here are the seven AI features worth your attention this week. Keep reading for that!

↩️ Don’t miss out: Miss our last newsletter? We covered: Thinking Machines released its first AI model, OpenAI unveiled the Codex Micro, and OpenAI introduced GPT-Red. And more. Check it here!

Ep 821: Claude Desktop Gets Upgrade, New Open Source Model Shocks, ChatGPT Desktop Gets Better and 7 More AI Features You Can Use Today

Is Kimi K3 the shocker of 2026?

Could be.

Now, we have a new (soon to be) Open Model that’s competing with Fable 5 and GPT-5.6, a feat few would have believed possible.

And that was the only new and important drop this week in AI.

Claude brought useful browser to the desktop, ChatGPT made a big fix to how ChatGPT Work works and Google rolled out avatars that could change content creation.

Don’t miss our Friday Features show, where we recap the most important AI updates and features you can use today.

Also on the pod today:

• Claude desktop adds built-in browser 🖥️
• ChatGPT desktop gets full sync 🔄 
• OpenAI’s universal search upgrade 🔍

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight –  River is An AI that demos and closes your deals, Graft AI gives AI agents a reliable way to use legacy software by turning real interfaces into stable tools, Cito is a Semantic search over academic papers.

Google Search AI — Now you can link your favorite apps like Instacart, Canva, and YouTube Music directly in Google Search's AI Mode.

Nemotron 3 Embed — NVIDIA just dropped Nemotron 3 Embed, their new open embedding models that set a retrieval benchmark record and offer flexible deployment for enterprise AI.

NVIDIA Cosmos 3 Edge — Nvidia’s new Cosmos 3 Edge model could spark the next big wave in robotics and physical AI.

Microsoft and 3M — Microsoft is teaming up with 3M to boost AI data center speed and reliability using new optical connectors

Google Vids Upgrades — Google Vids just got a boost with Gemini Omni and personal avatars, letting you create and edit videos using simple prompts, plus star in them with your own digital double.

China and AI — China’s Xi Jinping pushed for global teamwork on AI and warned about the risks of losing control.

OpenAI PR Chat — OpenAI's new PR Chat lets you review, edit, and accept pull requests without ever leaving Codex.

1Password and Claude — 1Password just launched a way for Claude to handle your logins without ever seeing your passwords.

Meta AI Safety — Meta wants AI to flag risky teen behavior on Instagram, but privacy experts say it could make things even messier.

Deepmind and Isomorphic Labs — Google DeepMind and Isomorphic Labs are teaming up to use AI for faster outbreak detection, safer drug design and stronger biosecurity.

Grok Automations — Automate your busywork with Grok: just describe the task once, pick when it runs, and let Grok handle it from there.

Roblox AI — Roblox is rolling out Build, a mobile-first AI tool that lets anyone create games from text prompts right in the app.

1. Nadella Faults Anthropic’s Fable Limits as Microsoft Pushes AI Choice 😱

His remarks arrive as Microsoft is pushing businesses toward cheaper, customizable AI models that can run on their own data instead of relying on a small group of major model providers.

2. Kimi K3 Ranks Third on Artificial Analysis, Open Weights Due July 27 🧑‍🎨

Kimi has launched K3, a 2.8-trillion-parameter multimodal model with a one-million-token context window, and says its full weights will be released by July 27.

According to Artificial Analysis, K3 scores 57, placing it third behind Claude Fable 5 and GPT-5.6 Sol, while outperforming several major rivals on agentic and knowledge-work tests.

3. EU Orders Google to Open Android and Search Data to AI Rivals 🔍

EU regulators have set new Digital Markets Act requirements that will force Google to give competing AI assistants access to 11 Android features and share certain anonymized search data with rivals such as OpenAI.

The Android changes are due with the next version in July 2027, while the search-data rules begin in January 2027, aiming to make it easier for alternative AI and search services to compete with Gemini and Google Search.

4. Google’s Gemini 3.5 Pro reportedly misses its June launch target 🎯

According to Bloomberg News, Google’s flagship Gemini 3.5 Pro model is running months late as the company works to strengthen its coding performance. The reported setback lands as OpenAI and Anthropic push competing systems, increasing pressure on Google to show that Gemini can keep pace at the high end of the market.


5. Meta’s Muse Spark 1.1 Arrives on OpenRouter for U.S. Developers ⚡

Meta says Muse Spark 1.1 is now available through OpenRouter, giving U.S.-based developers a quicker way to access the model without setting up a separate vendor relationship.

The move matters because OpenRouter acts as a single testing and billing hub for many AI models, putting Meta’s newest offering in front of a large developer audience immediately.

6. OpenAI Restores ChatGPT as the Main Screen After “Work” Backlash 👷

OpenAI has quickly revised its new desktop app after users complained that its core chat feature had been oddly sidelined by the launch of “ChatGPT Work.”

The updated app once again opens directly to the familiar “Message ChatGPT” interface, while Work, Codex, Projects, and recent chats have been moved into clearer secondary positions.

The priciest assumption in your AI strategy died late Thursday night: the one where frontier level intelligence only lives behind closed labs and $200 monthly subscriptions.

A Chinese AI lab kinda just tore that strategy down.

Moonshot AI dropped Kimi K3, a soon to be open weights monster punching up to trade blows with the two best closed models on the planet.

Meanwhile, Anthropic stuffed a full browser inside Claude Desktop, and OpenAI rebuilt ChatGPT Work after fumbling the first version.

Biiiig week.

Skip this one and you're prolly walking into Monday's AI strategy conversation armed with last week's intel.

Today's Everyday AI Friday Features show breaks down all seven updates, each one usable today, so you can be the smartest AI voice in the room by lunch. Let's get into it.

1. Claude Code Finally Browses the Web 🌐

Anthropic added a built-in browser to Claude Code desktop, letting it click, type, and act across external websites while handling a task. Paid macOS and Windows users get a clean browser profile, with approval required for every new site.

Finally.

Developers can keep documentation, bug trackers, dashboards, and execution inside one loop, while security teams retain explicit control over Claude’s external actions.

Try This

Choose one development task that requires documentation, a bug tracker, and a dashboard. Run the entire workflow inside Claude Code desktop and track the tab switching, copy-pasting, and manual handoffs eliminated.

2. ChatGPT Makes Your History Searchable 🔎

OpenAI added universal search across ChatGPT chats, Projects, images, and documents from one sidebar box. It’s rolling out to free and paid users across web, iOS, and Android.

Small? Only if your team enjoys rebuilding work it already created.

Old analyses, contracts, images, and unfinished thinking can now become reusable institutional knowledge instead of digital landfill, cutting the scrolling and expensive rework that punish heavy ChatGPT users.

Try This

Search for one valuable analysis, document, or image your team created months ago. Use it in a current project and compare the retrieval time against how long rebuilding that asset would have taken.

3. Superhuman Prewrites Emails Before You Open ✉️

Superhuman upgraded Auto Drafts to prepare responses and unanswered-email follow-ups before you open the inbox. Business and Enterprise users can set it up on desktop, with drafts syncing across mobile, Gmail, and Outlook.

Anthropic and OpenAI models replace GPT-3.5 underneath, which should make those drafts sound more like the actual user.

This won’t reinvent email, but shifting from writing every response to reviewing ready-made drafts can turn inbox management into a much faster approval workflow.

Try This

Use Auto Drafts for one workday and classify every suggestion as send, light edit, or complete rewrite. Keep it in the workflow only if the first two categories dominate and your average response time drops.

4. Spotify Turns Listening Into a Conversation 🎧

Spotify launched Talk to Spotify, a beta that lets Premium users control and discover content through an ongoing voice or text conversation. It’s rolling out on iOS and Android in the U.S., Ireland, and Sweden.

The advantage is your data.

Spotify already knows your playlists, repeat listens, and listening history, showing how apps with proprietary customer context can create experiences a general chatbot simply can’t copy.

Try This

Ask it to build a work playlist using your actual listening history, then refine the energy, genre, and recency through follow-up requests. Compare the result against manual playlist hunting based on setup time and how often you skip tracks.

5. Google Turns Your Likeness Into Video 🎬

Google added personal avatars powered by Gemini Omni, letting users record their face and voice once, then reuse that likeness in generated videos. It’s available to paid U.S. users age 18 or older and also works inside Google Vids.

The win isn’t replacing every real video.

It’s generating approved B-roll, training visuals, social clips, and executive footage when cameras or schedules become the bottleneck, turning one verified likeness into a reusable creative asset.

Try This

Choose one approved executive, trainer, or spokesperson and generate three short B-roll clips around an existing script. Compare production time, reshoots, and approval revisions against your normal video process.

6. ChatGPT Work Finally Feels Like ChatGPT 🔄

OpenAI updated ChatGPT Work desktop so chats and Projects appear in the sidebar, while Chat and Work history sync across web, mobile, and desktop. Users can switch modes more naturally, while local tasks remain on the computer.

The original setup felt disconnected.

Now teams can move between normal conversations and autonomous desktop work without rebuilding context or hunting through separate interfaces, removing the friction that quietly kills adoption.

Try This

Move one active Project from the web into the desktop app and alternate between Chat and Work while completing it. Count how often you repeat instructions, rebuild context, or hunt for a missing conversation.

7. Kimi K3 Blows Open the Model Race 🧠

Moonshot AI released Kimi K3, a 2.8 trillion-parameter model with native vision and a one-million-token context window. It’s live in Kimi, Kimi Work, and the API, with open weights scheduled for July 27.

Kimi K3 landed third on the Artificial Analysis Intelligence Index and beat closed flagships on several important benchmarks covered in today’s show.

Yeah, this is big.

Enterprise teams can evaluate frontier-adjacent autonomous work without accepting frontier pricing or vendor dependence, although Kimi K3 is far too large for normal consumer hardware.

Try This

Run the same long, multi-file business task through Kimi K3 and the closed model your team currently prefers. Compare accuracy, completion time, required interventions, and cost before brand familiarity picks the winner.

Reply

or to participate.