Good Morning! Here's what I have for you in today's newsletter:
Meta launches Muse Spark Open, its most capable open-weight model
OpenAI lets ChatGPT read and control your Chrome tabs
Anthropic makes Claude Code Review generally available for teams
The Opus 5.5 motion guide for people who don’t animate
4 new AI tools worth trying today
AI MODELS
Meta released Muse Spark Open, an open-weight version of Muse Spark built for coding and agent tasks. Meta calls it the most capable US open-weight model to date.
The model has 816 billion parameters but activates only 42 billion per token, using selected parts of its network to reduce the computing needed for each response.
A 1 million token context window gives it room for lengthy projects, with built-in support for text, images, video, and audio.
The release fulfills Mark Zuckerberg’s September promise to open Muse Spark, previously available only through the Meta Model API and the company’s own apps.
After charging for API access all summer, Meta is letting developers download the model itself. Running something this large still takes serious GPU hardware, so companies and research labs are the likely audience. For those equipped to host it, the appeal is straightforward: a capable coding model that keeps their source code on their own servers.
TOGETHER WITH FRONT
A fast AI answer isn't always a coordinated one. When AI resolves B2B issues without looping in the teams responsible for them, confidence outpaces alignment. A new Harvard Business Review Analytic Services briefing paper, sponsored by Front, looks at closing that gap.
AI AGENTS
OpenAI brought computer use to Chrome, letting ChatGPT read open tabs in Chat and operate them in Work to send messages, fill out forms, and download files.
In Chat, ask about pages you already have open, and ChatGPT can pull information from across those tabs to answer.
In Work, it can carry out browser tasks, including completing forms, sending messages, and saving files.
Too many tabs to keep track of? Ask ChatGPT to sort them into groups so related pages stay together.
Expense reports and copying details between browser tabs are exactly the chores people would happily hand off. Work users can now do that inside Chrome. Because ChatGPT uses your signed-in accounts, its actions have real consequences, including messages sent and forms submitted. Start with simple tasks on pages that don’t hold sensitive information.
AI TOOLS
Anthropic made Code Review generally available on Claude Team and Enterprise plans, with faster reviews, better detection of significant bugs, and settings that let admins choose how closely it checks each repository.
Choose Quick, Standard, or Deep reviews for each repository, giving critical code more scrutiny without applying the same settings everywhere.
Anthropic says the updated reviews run faster and catch more consequential bugs, reducing the wait for useful feedback.
Every Team and Enterprise organization can now access the feature, with review settings managed by admins at the repository level.
A finished pull request can spend hours waiting for someone to look at it. Claude can give changes a first pass, flagging potential bugs and missed edge cases before a teammate gets involved. That leaves human reviewers more time for design decisions. Deep reviews are an option for repositories where a mistake could cause an expensive outage.
HOW TO AI
Give Claude a clear brief, refine the frames, and export a loop for X.
Opus 5.5 can write an animation and render it inside the Claude app without installing tools on your computer. You'll need access to the model with code execution enabled.
Step 1: Test the renderer
Start a new Opus 5.5 chat. Ask for a two-second MP4 of a growing circle, built from one self-contained HTML file. If code execution isn't available, check Settings, then Capabilities. Tell Claude not to load fonts or scripts from outside websites.
Step 2: Write the brief
Use eight lines: GOAL, CANVAS, TIMING, LOOK, MOTION, SHOTS, AVOID, and CHECKS. Specify hex colors, text sizes, frame counts, and one main movement per shot. Use fonts available in the workspace. "Make it premium" leaves Claude to guess; a 96 px name on a #F7F4EF background doesn't. List the effects you never want under AVOID, so they don't return in later drafts.
Step 3: Render the loop
Request five named still frames first. Check the text, colors, and opening and closing positions. Once they look right, ask for the MP4 and a contact sheet showing every 15th frame. Watch the video several times, and use the contact sheet to find the frame number of anything you want changed.
Step 4: Refine the frames
Give a frame number and a specific change: "At frame 100, scale the name to 1.08 and back over six frames." For an unexplained jump, ask Claude to compare nearby frames before editing. Send one or two changes per round, so you can see whether each fix worked.
Step 5: Export for X
Request an H.264 MP4 with yuv420p and a GIF using one palette. Have Claude repeat the MP4 three times in a preview, then watch the transitions yourself before uploading. Check that the loop makes sense without sound, and preview it on your phone before you post.
P.S. You can access all my AI courses, trainings, prompts library, and $500+ worth of AI tool discounts here.
OpenAI rolled out GPT-Live-1.1 to Voice in ChatGPT, Codex, and dots for all users, adding new voices in French, German, Hindi, Mexican Spanish, Korean, and Japanese, better accents, and fewer pauses while it checks, with API access coming soon.
NVIDIA released Nemotron 3.5 Super VL, an open model with 120 billion total and 12 billion active parameters that combines reasoning, tool use, and image, document, and video understanding, also available in Perplexity for Pro and Max users.
Firecrawl launched Fireround, company funding intelligence in ChatGPT and Claude that tracks new rounds as they are announced, covers over $10 billion in weekly deals, and researches 8,000+ investors, free through Alexandria.
🖼️ Krea: opened the Krea 3 beta, an image editor that makes end-to-end 2K edits in under 8 seconds with bounding-box control and multi-turn edits.
🚨 Devin: launched Devin Oncall, which investigates alerts and bugs automatically to help teams diagnose and resolve production incidents.
🔁 Manus: can now run repetitive work on a schedule, starting at a set time and delivering the finished result you describe.
🎬 HyperFrames Studio: added Reference Any Video for recreating any video format, plus brand kits that pull colors, fonts, and logos from your site.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial
✓ $500+ worth of AI tools