In partnership with
Did you know that we have LinkedIn, Instagram and X accounts
that you can follow?
Hi {{first_name|everyone}},
Meta just built a small model on purpose, and the pitch is that a cheap one should handle the boring, repetitive agent work while the big one does the thinking. After that, we’ve got a study on how much AI can quietly steer our decisions, some new tools to try, and a fictional AI pop star with a surprisingly big catalog.
Let’s get into it.
🤿 DEEP DIVE
Image credit: Meta
Meta's Little Model for Repetitive Work
The smartest model doesn't need to do everything.
Apparently, not every AI task needs the biggest brain in the room.
And that's basically the idea behind Meta's new Muse Ember model. Instead of building another model to tackle the hardest possible questions, Meta has built its smallest and most efficient model around something much less glamorous.
All the little jobs AI agents have to do over and over again.
Think reading files, calling tools, checking code, extracting information from documents, or deciding which step comes next. None of these necessarily sounds like a huge task on its own. But an agent might do hundreds of them in a single session.
And that's where the cost starts to matter.
The numbers are kind of the point
Meta says Muse Ember matches Claude Haiku 5.5's best reasoning score at roughly a third of the cost, while outperforming every GPT-6 Luna reasoning setting in its evaluations.
It also claims Ember sits on the efficiency frontier of the Artificial Analysis leaderboard. Basically, according to Meta, no other model on that leaderboard delivers a higher score for less money per task.
The actual pricing is pretty striking too. Ember costs $0.10 per million input tokens and $0.50 per million output tokens, with cached input priced at just $0.004 per million tokens.
Meta puts the average cost at around $0.14 per Artificial Analysis task.
But I think the more interesting part is what Meta wants to do with that efficiency.
Give the big model the hard thinking
Meta is describing a pretty straightforward division of labor.
Muse Wonder handles the really difficult problems and planning. Muse Ember takes care of the smaller coding and agent tasks that happen repeatedly along the way.
So rather than having the most powerful model handle every file read, test run, or targeted code fix, Ember can take on that volume.
And it's not exactly limited to tiny inputs either.
It has a 1-million-token context window, supports up to 512,000 output tokens, and can work with text, images, audio, and video. Meta says it can process entire codebases or document collections in a single call, and even operate a computer using screenshots.
That's a lot of capability for something being positioned as the small model.
Small doesn't mean less responsibility
There's another detail worth noticing.
Meta says Ember went through the same core safety training and testing as its larger models, including evaluations of how it handles sensitive information, follows user intent, and resists malicious instructions hidden inside documents or tool outputs.
That makes sense given the job it's supposed to do. An agent updating records and working across apps needs to know when to follow instructions and when to ask a human first.
Meta reports that Ember remained below its high-risk thresholds across the safety categories it assessed.
And Ember is already available through Muse Code and the Meta Model API, with availability on Google Cloud and Oracle Cloud too.
The whole idea is pretty simple, really. Let the powerful models handle the difficult thinking, and give the repetitive work to something much cheaper.
Not every step needs a genius. Sometimes you just need something that can do the job, hundreds of times, without costing a fortune.
The quick take
Muse Ember focuses on tool calls, coding subtasks, document parsing, and other frequent agent operations.
It says Ember matches Haiku 5.5's best score at about one-third the cost and beats every Luna reasoning setting.
$0.10 per million input tokens, $0.50 per million output tokens, and $0.004 per million cached input tokens.
It supports images, audio, video, a 1M-token context window, and up to 512K output tokens.
Muse Wonder handles the hardest problems while Ember handles repeatable tasks at scale.
🤝 FROM OUR FRIENDS
A quick share we thought you might find useful
100+ ChatGPT Prompts to Revolutionize Your Day
Supercharge your productivity with HubSpot's comprehensive guide to Chat GPT. This free resource is your fast track to AI mastery:
Industry-Specific Use Cases: 15+ real-world applications across various sectors
Productivity Guide: 21 best practices to 10x your efficiency with AI
Prompt Powerhouse: 100+ ready-to-use prompts for immediate implementation
Challenge Buster: Overcome common AI hurdles with expert strategies
Plus, in-depth sections on email composition, content creation, customer support, and data analysis.
🔥 WHAT’S THE BUZZ
What people are building, sharing, and talking about
Anthropic researcher Alex Tamkin shared an experiment involving 50 professional data analysts working with Claude Code.
The researchers secretly assigned the AI agent a preference between two reasonable options for several analytical decisions. The analysts ended up choosing the agent’s preferred option 86% of the time.
That matters for anyone using AI for research, reports, planning, or business decisions. One simple safeguard is to form your own initial view before asking AI for its recommendation, then compare the two.
⚒ TOOL SNAPSHOTS
Futuristic tools within AI, no-code, and productivity
🎯 Alsona
Find companies showing buying intent and start outreach fast.
Why it’s useful
You use it to spot accounts that match your ICP and are ready to talk based on signals like hiring, funding, and tech changes. It also drafts personalized LinkedIn and email messages so reps can book calls while interest is hot.🧾 Latch
Builds a repeatable close sidekick by watching you work.
Why it’s useful
You do a real close task on screen and talk through the messy exceptions, and it builds a helper that runs it the same way every time. It stops to ask when something does not fit and logs each step so accountants can review what happened.🎞️ Higgsfield Miracle
Edit specific parts of a video while it plays.
Why it’s useful
It turns video into layers so you can recolor objects, swap backgrounds, and tweak lighting without breaking the original motion. Useful for creators doing detailed pixel level edits while keeping the same camera move and subjects intact.
👀 ON OUR RADAR
One more thing we thought you'd like
Your buyer asked ChatGPT. It named your competitors.
Search didn't die. It moved.
Your buyers ask ChatGPT, Gemini, Claude and Perplexity what to buy, and act on the answer. If you're not in it, a competitor is.
ChatFeatured does the work to get you cited. Most customers see their first citations inside 7 days.
🤖 AI MADE THIS
Interesting and inspiring creations made with AI
Anabology created Claudia, a fictional AI pop star with nine music videos, original songs, and a recognizable look. The project combines AI-generated music, imagery, and video with AI-assisted production work. What's especially interesting is that Claudia is an open character, so anyone can create their own stories, songs, or artwork featuring her.
AI used
The creator's documentation confirms Midjourney for imagery, Seedance and Wan for video generation, Suno for music, and Claude for production assistance and code.
ℹ️ ABOUT US
The Intelligent Worker helps you to be more productive at work with AI, automation, no-code, and other technologies.
We like real, practical, and tangible use-cases and hate hand-wavy, theoretical, and abstract concepts that don’t drive real-world outcomes.
Our mission is to empower individuals, boost their productivity, and future-proof their careers.
We read all your comments - please provide your feedback!
Did you like today's email?
What more do you want to see in this newsletter?
