Glimpse.
Matt Wolfe
23 min video
3 min read
Claude's Browser, Spotify AI, and OpenAI's First Hardware
You just saved 20 min.
The big takeaway
Claude gets an in-app browser for code editing, Spotify launches AI chat for music discovery, Google expands AI search with app integrations, new open-weight models emerge (Kimi K3 most capable), and OpenAI releases its first hardware: a specialized keyboard for Codex. Plus: Siri upgrade, Meta rolls back AI image tagging, and distributed home-based AI compute pilots launch.
Claude & Code Editing
Claude Code Gets In-App Browser
Claude's desktop app now includes a built-in browser accessible via Cmd+Shift+B or a browser icon in Code projects. Users can annotate webpage elements, highlight them, and have Claude modify corresponding code—useful for live website editing and data scraping without API costs.
Browser Replaces Expensive APIs
Instead of using costly X (Twitter) or Instagram APIs to pull data, Claude can now browse these sites directly and extract information, significantly reducing operational costs for data collection tasks.
Google Search & Vids Evolution
AI Search Connects Apps On-Demand
Google's AI search mode now intelligently connects third-party apps based on user queries. Ask for a grocery list and it offers Instacart integration; the system detects relevant apps and prompts connection without a traditional app store interface.
Google Vids Adds Animation & Avatars
Google Vids now integrates Gemini Omni to animate slide content, supports chat-based video editing, and lets users upload personal avatars to appear in generated videos—similar to features in Sora and Meta products.
Spotify's Conversational AI
Talk to Spotify for Music Discovery
Premium Spotify users can now chat or speak directly in the app to ask about music, revisit listening history, create playlists, and explore podcasts and audiobooks. The AI understands context (e.g., 'What was I listening to 2 years ago?') and generates relevant playlists without leaving the app.
New Open-Weight Models
Inkling Model from Thinking Machines Labs
Mera Morati's new lab released Inkling, a 95-billion-parameter open-weight model available on Hugging Face. Performance is solid for an open model but trails GLM-4-2 and closed models on most benchmarks; playable free on Tinker playground.
1
GLM-4-2
Best open-weight
2
Inkling
Decent open-weight
3
Closed models (GPT-4o, Claude 3.5, etc.)
Superior overall
Open-weight model performance hierarchy
Bonsai 27B: Phone-Sized AI
Prism ML's Bonsai is a 1-bit quantized, 27-billion-parameter model compressed to 4GB that runs entirely on-device on phones. Fast (94 tokens/sec) but struggles with coding; designed for quick answers, not complex reasoning.
27B parameters
Largest model for phones
Bonsai 27B model size
Kimi K3: Most Capable Open Model Coming
Moonshot AI's Kimi K3 (2.8 trillion parameters, weights opening this month) outperforms GPT-4.5, Claude Opus 4.8, and GLM-4-2 on most benchmarks, matching or beating Claude 3.5 Sonnet and o1 on coding and reasoning tasks. First open model to reach 2.8T parameters.
Kimi K3 (open, coming)
2.8 T parameters
Inkling
0.095 B parameters
Bonsai 27B
0.027 B parameters
Model parameter sizes compared
AI Agent & App Updates
Grok Gets Automations & Open Source
Grok added automation triggers (daily schedules, email inbox watches) similar to Claude and ChatGPT, and open-sourced Grok Build, its coding agent and terminal UI, so developers can fork and extend it.
ChatGPT Improves Search Across Platforms
ChatGPT enhanced search on web, iOS, and Android to find past conversations, projects, images, and documents in one place with filtering—making it easier to retrieve prior work.
Claude Accesses 1Password Credentials
Claude can now use 1Password to auto-login to websites without exposing passwords to Claude itself—credentials stay encrypted in 1Password and are never sent to or trained on by Anthropic.
DoorDash CLI for AI Agents
DoorDash released a command-line interface so any AI agent (OpenClaw, Hermes, Codeex) can order food directly—search stores, find deals, and checkout. Currently on waitlist for US and Canada.
Meta, Apple & Platform News
Meta Rolls Back AI Image Tagging Feature
Meta's Instagram feature allowing users to generate AI images of tagged people (from last week) faced immediate backlash from creators and actors and was rolled back within a week.
New Siri Rolls Out with AI Features
Apple's updated Siri now includes AI capabilities on iPhone and Apple Watch, with early feedback positive. Apple Watch users now get useful AI features on-device, not just voice commands.
Google Images Becomes Pinterest-Like
Google is redesigning Google Images to learn user preferences, suggest images, and add in-search image generation—rolling out soon to google.com as part of 25 years of visual search celebration.
Notebook LM Rebrands to Gemini Notebook
Google is renaming Notebook LM to Gemini Notebook; same product and functionality, but the name change may briefly confuse existing users.
OpenAI Hardware & Legal
OpenAI Launches First Hardware: Codex Keyboard
OpenAI released the Codex Creator Micro keyboard (co-designed with Work Louder) for $230 at openai.com/supply. Specialized for Codex use; available in clicky and non-clicky versions.
$230
Codex Creator Micro price
OpenAI's first hardware product
Screenless Smart Speaker Rumored
Unconfirmed reports suggest OpenAI is developing a mobile, screenless smart speaker (like Amazon Alexa) that acts as a humanlike AI companion, controls smart home devices, and integrates ChatGPT—possibly inspired by Apple hardware ideas.
Apple Sues OpenAI Over Trade Secrets
Apple filed a lawsuit claiming former Apple employees brought trade secrets and hardware examples to OpenAI. OpenAI denies the allegations. Rumors suggest the new device may incorporate ideas originally from Apple.
Infrastructure & Policy
New York Enacts Data Center Moratorium
New York became the first state to block new hyperscale data centers over 50 megawatts for up to a year, giving regulators time to address energy costs and environmental impact concerns.
1 year
NY data center moratorium duration
First state-level AI infrastructure pause
Home-Based Distributed AI Compute Pilot
SunRun is launching a pilot program placing mini compute nodes (with solar and battery storage) in customers' homes. Homeowners get paid for sharing compute capacity with AI companies—a decentralized alternative to massive data centers.
Other Notable Updates
Anthropic Launches Claude for Teachers
Verified K-12 educators in the US get free premium Claude access, teaching skills library, and curriculum connections—reducing cost barriers for educational use.
Seedream 5.0 Pro Multimodal Generator
ByteDance's Seedream 5.0 Pro (available on Artlist) generates images from text and up to five reference images to control characters, colors, composition, and style. Stronger reasoning enables detailed infographics, UI mockups, and readable text in images.
Worth quoting
"That's a cool idea. Let's add it into our app."
— Matt Wolfe (paraphrasing Anthropic's reaction to ChatGPT browser), at [0:31]
"It's probably a really bad idea cuz everybody's going to judge me."
— Matt Wolfe, at [4:38]
"There's never a dull week in AI, and this roller coaster just keeps on moving forward."
— Matt Wolfe, at [22:05]
Made with Glimpse by Wozart
glimpse.wozart.com/v/bfn67w01
Share this infographic
Read this infographic as text

Claude's Browser, Spotify AI, and OpenAI's First Hardware

Summary of the video “AI News: Claude's New Browser, Spotify Gets AI & OpenAI's New Hardware” by Matt Wolfe.

Claude gets an in-app browser for code editing, Spotify launches AI chat for music discovery, Google expands AI search with app integrations, new open-weight models emerge (Kimi K3 most capable), and OpenAI releases its first hardware: a specialized keyboard for Codex. Plus: Siri upgrade, Meta rolls back AI image tagging, and distributed home-based AI compute pilots launch.

Claude & Code Editing

Claude Code Gets In-App Browser

Claude's desktop app now includes a built-in browser accessible via Cmd+Shift+B or a browser icon in Code projects. Users can annotate webpage elements, highlight them, and have Claude modify corresponding code—useful for live website editing and data scraping without API costs.

Browser Replaces Expensive APIs

Instead of using costly X (Twitter) or Instagram APIs to pull data, Claude can now browse these sites directly and extract information, significantly reducing operational costs for data collection tasks.

Google Search & Vids Evolution

AI Search Connects Apps On-Demand

Google's AI search mode now intelligently connects third-party apps based on user queries. Ask for a grocery list and it offers Instacart integration; the system detects relevant apps and prompts connection without a traditional app store interface.

Google Vids Adds Animation & Avatars

Google Vids now integrates Gemini Omni to animate slide content, supports chat-based video editing, and lets users upload personal avatars to appear in generated videos—similar to features in Sora and Meta products.

Spotify's Conversational AI

Talk to Spotify for Music Discovery

Premium Spotify users can now chat or speak directly in the app to ask about music, revisit listening history, create playlists, and explore podcasts and audiobooks. The AI understands context (e.g., 'What was I listening to 2 years ago?') and generates relevant playlists without leaving the app.

New Open-Weight Models

Inkling Model from Thinking Machines Labs

Mera Morati's new lab released Inkling, a 95-billion-parameter open-weight model available on Hugging Face. Performance is solid for an open model but trails GLM-4-2 and closed models on most benchmarks; playable free on Tinker playground.

Bonsai 27B: Phone-Sized AI

Prism ML's Bonsai is a 1-bit quantized, 27-billion-parameter model compressed to 4GB that runs entirely on-device on phones. Fast (94 tokens/sec) but struggles with coding; designed for quick answers, not complex reasoning.

Kimi K3: Most Capable Open Model Coming

Moonshot AI's Kimi K3 (2.8 trillion parameters, weights opening this month) outperforms GPT-4.5, Claude Opus 4.8, and GLM-4-2 on most benchmarks, matching or beating Claude 3.5 Sonnet and o1 on coding and reasoning tasks. First open model to reach 2.8T parameters.

AI Agent & App Updates

Grok Gets Automations & Open Source

Grok added automation triggers (daily schedules, email inbox watches) similar to Claude and ChatGPT, and open-sourced Grok Build, its coding agent and terminal UI, so developers can fork and extend it.

ChatGPT Improves Search Across Platforms

ChatGPT enhanced search on web, iOS, and Android to find past conversations, projects, images, and documents in one place with filtering—making it easier to retrieve prior work.

Claude Accesses 1Password Credentials

Claude can now use 1Password to auto-login to websites without exposing passwords to Claude itself—credentials stay encrypted in 1Password and are never sent to or trained on by Anthropic.

DoorDash CLI for AI Agents

DoorDash released a command-line interface so any AI agent (OpenClaw, Hermes, Codeex) can order food directly—search stores, find deals, and checkout. Currently on waitlist for US and Canada.

Meta, Apple & Platform News

Meta Rolls Back AI Image Tagging Feature

Meta's Instagram feature allowing users to generate AI images of tagged people (from last week) faced immediate backlash from creators and actors and was rolled back within a week.

New Siri Rolls Out with AI Features

Apple's updated Siri now includes AI capabilities on iPhone and Apple Watch, with early feedback positive. Apple Watch users now get useful AI features on-device, not just voice commands.

Google Images Becomes Pinterest-Like

Google is redesigning Google Images to learn user preferences, suggest images, and add in-search image generation—rolling out soon to google.com as part of 25 years of visual search celebration.

Notebook LM Rebrands to Gemini Notebook

Google is renaming Notebook LM to Gemini Notebook; same product and functionality, but the name change may briefly confuse existing users.

OpenAI Hardware & Legal

OpenAI Launches First Hardware: Codex Keyboard

OpenAI released the Codex Creator Micro keyboard (co-designed with Work Louder) for $230 at openai.com/supply. Specialized for Codex use; available in clicky and non-clicky versions.

Screenless Smart Speaker Rumored

Unconfirmed reports suggest OpenAI is developing a mobile, screenless smart speaker (like Amazon Alexa) that acts as a humanlike AI companion, controls smart home devices, and integrates ChatGPT—possibly inspired by Apple hardware ideas.

Apple Sues OpenAI Over Trade Secrets

Apple filed a lawsuit claiming former Apple employees brought trade secrets and hardware examples to OpenAI. OpenAI denies the allegations. Rumors suggest the new device may incorporate ideas originally from Apple.

Infrastructure & Policy

New York Enacts Data Center Moratorium

New York became the first state to block new hyperscale data centers over 50 megawatts for up to a year, giving regulators time to address energy costs and environmental impact concerns.

Home-Based Distributed AI Compute Pilot

SunRun is launching a pilot program placing mini compute nodes (with solar and battery storage) in customers' homes. Homeowners get paid for sharing compute capacity with AI companies—a decentralized alternative to massive data centers.

Other Notable Updates

Anthropic Launches Claude for Teachers

Verified K-12 educators in the US get free premium Claude access, teaching skills library, and curriculum connections—reducing cost barriers for educational use.

Seedream 5.0 Pro Multimodal Generator

ByteDance's Seedream 5.0 Pro (available on Artlist) generates images from text and up to five reference images to control characters, colors, composition, and style. Stronger reasoning enables detailed infographics, UI mockups, and readable text in images.

Notable quotes

That's a cool idea. Let's add it into our app. — Matt Wolfe (paraphrasing Anthropic's reaction to ChatGPT browser)
It's probably a really bad idea cuz everybody's going to judge me. — Matt Wolfe
There's never a dull week in AI, and this roller coaster just keeps on moving forward. — Matt Wolfe

More like this