@matthewbermanOct 26

Muse Wonder Full is slow.

@theoOct 23

I still think you guys are severely underrating Muse Wonder Full. Been using it a ton for orchestration and drafting rewrites of docs, skills, mcp descriptions and more. It's so much more pleasant to talk to AND it's great at steering models (as such, decent at skill writing!)

@exm7777Oct 18

every week a new "frontier" model drops and everyone reads benchmarks... but the only thing that matters is: where does it fit YOUR work? here's what i do: > run my real tasks through it > find 3-5 spots where it's significantly better than my current model > push it until it breaks that's how i figured out i'll never be using Muse Wonder, even if it's cracked for specific coding workflows

@alexandr_wangOct 16

Known affiliation

after an exciting week of launches (muse wonder [full] & muse spark open), i want to write a note of appreciation to the amazing research teams at meta superintelligence labs on reaching such a special milestone ♥️ we started MSL and TBD a little more than a year ago and set a goal of producing a frontier model within a year or so. given where llama4 landed, that was an incredibly audacious goal, and we knew it wouldn't be easy. many people told us it would be impossible. we knew it was unlikely, but still worth doing. a remarkable group of researchers and engineers joined the effort from inside and outside meta. every single person came in, picked up a shovel, and got to work with earnestness and urgency. nobody asked whose job it was. we assembled a team of the most brilliant people i have ever had the opportunity to work with (and remember, i was an HRT intern 🙃). i am surrounded by ambitious, visionary, technically excellent, hard-working, pragmatic, thoughtful scientists, engineers, and product folks—people who will chase a weird loss spike past midnight just because they need to know why. these models are the product of their passion, hard work, and brilliance. somewhat paradoxically, when we started TBD, we set a principle of no deadlines. we knew we had an astronomical amount of work ahead of us—across pretraining, post-training, RL, multimodal, infrastructure, safety, data, AND frontier research. there was no use setting arbitrary timelines—we just had to get to work, be thoughtful, and build the foundations of a frontier research effort in record time. and run after run, this team did exactly that. our researchers are magicians. i can only describe building these models with this team as a wondrous journey of scientific discovery. i am constantly amazed by what we can accomplish with scaling. i will remember this past year for the rest of my life—the late nights, the broken runs, the evals that came back better than any of us dared to hope. watching this team create wonder after wonder has been a true and deep gift. more than just the models we released this week, we have built a machine that builds new models. it often sounds like a cop-out to say labs can just turn the crank to keep scaling, but now i have seen it first-hand. sure, it's a lot of work and a careful process, but we have built a scalable system to keep advancing at the frontier, and i am really proud of how fast this team keeps making these models better. lastly, i get to work with researchers who care deeply about how their work benefits humanity. it's easy to forget when everything is changing so fast, but the whole damn point is to make people's lives better. time and time again, i have seen our researchers, engineers, and product folks build our technology with care so it can truly be good for the world—whether that's building the best models for muse, thoughtfully open-sourcing strong models, or making sure the alignment and security of our systems are rock solid. this is a team of people who never forget there's a real person on the other side of every prompt. thank you to everyone at MSL for being part of this crazy journey over the past year. i'm so grateful for every one of you, and our work is not done!

@aakashguptaOct 16

Six months ago, Meta wouldn't let anyone outside the company download this AI. This morning they put it online for free so literally anyone can grab it, and they did it three days after building a better one. Pay attention to that order of events, because it's how Meta decides what to give away now. Back in April, Meta called Muse Spark the most powerful model it had ever made and kept the weights locked up. On Monday they launched Muse Wonder, which beats it. This morning all 816 billion parameters of Muse Spark went up on Hugging Face. So the rule at Meta seems pretty simple at this point. Something goes open source the week it stops being the best thing Meta sells. That's a big change from the Llama days. When Zuck released Llama 3.1 405B in 2024, it was the best model Meta had, and he wrote a whole letter about why open source was the future. Meta didn't sell AI back then, so giving it away cost them nothing and made OpenAI's product a little less special every time someone downloaded it. Fast forward to this week and Meta has an actual AI business. Muse Wonder runs $4 per million input tokens and $20 per million output, about what Anthropic charges for Claude Opus 5.5, and the top version, Wonder Full, isn't on the API at all. You have to pay for a Muse Code subscription to touch it. Now look at what they gave away. Meta's own benchmark chart has Kimi K3 beating Muse Spark Open on Terminal-Bench 2.1, 85.0 to 76.4, and GLM 5.3 and Qwen beat it there too. That's why the announcement says "most capable US open-weight model," with the US in there on purpose. Think about who actually wants this thing. It's the American company whose lawyers won't approve Chinese models and who's paying OpenAI or Anthropic for a mid-range model right now. Meta just handed that company a free version, which eats into the cheaper tier where its rivals make a lot of their money, and Meta's own paid model sits safely one step above it. Then there's Apple. The whole model runs on two Mac Studios with 512GB of memory each, a setup Apple won't even start selling until later this month. Zuck thanked Apple CEO John Ternus by name. In 2022, Apple's privacy changes cost Meta about $10 billion in ad revenue in one year. Both companies come out ahead when developers run AI on their own desks instead of renting it from OpenAI or Anthropic, and apparently that's enough to get two old rivals to share a launch day. My bet is Muse Wonder's weights show up right around the week Meta ships something that beats it.

@itsolelehmannOct 15

American open source is FLYING HIGH again 🇺🇸 Zuck this week: ships Muse Wonder, open-sources Muse Spark, and gets Apple to tweak its own software so the thing runs well on a Mac 18 months ago people were dunking on Llama 4's benchmarks. now Meta's open-source model beats Kimi K3, GLM 5.3 and Qwen on agent tool use and the hardest reasoning test (HLE). what a turnaround story Meta is right now

@harrystebbingsOct 15

Why Meta's new model is a zero “It is a zero because there is no API. We cannot even use it. It is only for internal use by Meta. This is the most performative eval I have seen to date. I am not saying it is not nominally true, and clearly the team has done good work. What is the point of a model that allegedly has parity for the same price, is not even cheaper, and you cannot use?” @jasonlk Love to hear your thoughts @grmcameron @alexatallah @kavyakvk @simonw

@harrystebbingsOct 15

For the last 5 decades straight, we have created (as a society) circa 20 decacorns per decade. In the last 12 months, we have created 42. The world is changing faster than ever before and if you do not stay up to date. You will be left behind. That is why I created this show with Jason and Rory. This is the only show in tech, you have to listen to every week. No egos, no politics, just tech. On the agenda: - Valar Atomics Sues Day One Ventures - Meta Returns to the AI Frontier With Muse Wonder - NVIDIA Eyes $30B Reflection Deal - Jev Raises $870M at a $7.5B Valuation My notes below with @jasonlk and @rodriscoll 1. We Generated 20 $10B Companies per Decade for the Last 5 Decades and Now 42 in a Year Tech produced roughly 20 decacorns per decade for five decades, yet 42 have been created in less than a year of this decade. Only about 130 tech companies have ever been worth $10B or more. Rising market caps explain part of the jump, but the pace is stunning. 2. Who Is Right and Who Is Wrong: Day One or Valar? X wrongly claimed Day One was suing its portfolio company, when Valar Atomics wants a declaratory judgment to move on. Waiving pro rata for Sequoia is common, but letting an existing investor in as a strategic is playing favorites. Still, once founders decide, early investors must get with the program because it is not their company. 3. Why Meta's New Model Is a Zero Meta's new model scores well on evals, but there is no API and it can only be used inside Muse. A model claiming parity at the same price, not cheaper and not usable by developers, is the height of performative evals. 4. Why We Should Lead Jev's Round at 8 Billion Jev is a super fast, super cheap classifier, accurate enough for maybe 10% of AI use cases. Cloudflare and Amazon can match it today, but the bet is that a focused team out of OpenAI pulls ahead while those efforts stay niche. If Jev becomes hyper accurate, it changes the game. 5. What Is the Single Greatest Potential for AI? Drug discovery is probably AI's largest opportunity, with drugs that dramatically extend longevity and quality of life. If AI companies can accelerate discovery and own the IP while others do the hard work, AI could reboot the industry. Trials would not speed up, but the prize is a GLP-1 out five years faster. 6. Haiku 5.5 Is the First Time AI Has Been Deflationary Falling token prices have been offset by consuming far more tokens, so AI costs have not fallen. Haiku 5.5 is almost as good as Sonnet for many use cases at a tenth of the price, and Sonnet 5.5 is twice as fast. After a 90% price cut, usage must rise 10 times for revenue to hold. 7. Why the SaaSpocalypse Is Worse Than We Thought The SaaSpocalypse passed only for companies well above $1B in revenue, and everyone without massive scale is living a worse one than in March or April. Leaders will acquire some companies and grind out the rest. A safe software IPO may now need $1.5B to $2B of ARR rather than $500M. (links in comments)

@aakashguptaOct 15

Mark Zuckerberg was about to spend up to $200 billion, more than the entire Disney company is worth, on an AI that didn't exist yet. So he asked the guy building it if he was sure it would work, and even that guy wasn't sure. The guy is Alexandr Wang, who runs Meta's AI lab, and the story came out on John Collison's podcast this week. Late last year Meta's top executives were together in Hawaii, and Wang says the CFO, Susan Li, pulled him aside more than once on that trip to warn him that if the model flopped, it would be really, really bad for Meta's finances. All he could show her was his gut read on how the research was going and a few charts. She had every reason to be nervous. Meta spent $72 billion on capex in 2025, and the plan for 2026 now runs $130 to $145 billion. Last quarter, free cash flow at one of the most profitable companies on the planet fell to $784 million, and Meta borrowed about $25 billion in the bond market in the same stretch. So why not wait until you know it works? Because in AI you have to pour the concrete years before you find out what it's for. Meta started planning its two giant clusters, Prometheus in Ohio and the 5 gigawatt Hyperion in Louisiana, back in 2023 and 2024. That was before Llama 4 disappointed, and before Wang ever got the WhatsApp message from Zuck asking if he had time for a call. Meta was building the biggest computers in its history for an AI lab that didn't exist yet. Wang called the whole thing the opposite of Lean Startup, where you test something small and only spend real money once it works. Muse Wonder got planned in August of last year and shipped this Monday, and the biggest checks were signed with the answer still most of a year away. What makes it even more nerve-wracking is how few people the bet rides on. Zuck says the research team is under 200 people, closer to 100, with infrastructure staff on top. Run his own numbers and every researcher in that lab is carrying something like a billion dollars of Meta's money. When John Collison asked whether Meta will actually make money from any of this, Zuck's first answer was "The answer better be yes." A second later he corrected himself and just said yes.

@signulllOct 13

i am not a model reviewer, nor am i an influencer but having used muse wonder for few days now, i can def say it’s a reliable & capable frontier model more so than any other meta model making endeavor thus far. i enjoyed its ability to focus on very personal tasks as well as the long term memory layer stuff (although i need a lot more time with this). i think in the macro it’s interesting how vertical some of this stuff might really get when you specifically design models to power very specific types of experiences, in muses case obviously that is deeply personal intelligence more so than work or even business critical coding (although the model is very capable of that too). this is even more true because some personal startups are built off of fine tuned open source models too further amplifying the vertical narrative. as the software app layer gets commoditized, is the model effectively going to be the actual product in some sense? that’s also a huge possibility at this point. would’ve been baller if they just made it open weight though, imagine the ripple effects that would’ve caused for the american open weight eco system? regardless, i would’ve loved to be in the room when the open weight discussion happened, can you imagine hearing the perspectives on that decision?

@alexandr_wangOct 13

Known affiliation

Cheeky Muse* is back! Agrippa Zuckerberg @finkd and Euler Wang @alexandr_wang came to the pub to discuss their new frontier model, Muse Wonder. * made with Muse Wonder 🤝 00:00:24 Muse’s breakout use cases 00:07:28 The model is the product 00:14:35 Muse Wonder 00:24:56 How frontier models are built 00:48:26 Betting $100B 01:07:21 Agentic commerce 01:19:34 OpenRouter 01:20:06 Beyond the smartphone 01:27:00 Funding 01:34:06 Why AI won’t be winner-take-all 01:40:46 Meta Superintelligence Labs (MSL) 02:08:11 Working for Mark Zuckerberg

@rauchgOct 13

Congrats Gemini team. Never bet against Google, the company that brought us Transformers and has overwhelming data, engineering talent, and compute. Everyone has, rightfully, pivoted back to coding. Coding agents are the killer app of AI. We're firmly at an elysium for developers and builders. Incidentally, getting Muse Wonder and Gemini Argon back-to-back is a reminder why you as an individual, and your company, should stay model agnostic and gateway-maxxx / fx-maxxx.

@exm7777Oct 13

i spent most of my day working with Muse Wonder... here's my honest take on this model: - it thinks and approaches problems VERY differently than Opus or Sol - the vibe of talking with it is great, it doesn't sound like any other model (clearly not distilled from Claude lol) - it's excellent for long runs, basically any SWE work - but outside of that it's really unbalanced: writing is meh, anything 3D is far behind, near zero taste or creativity it feels like this model was trained purely to dominate agentic work and intense coding sessions, and it falls behind in every other department for me i don't think i'll use it much beyond that... Opus 5.5 is way more versatile and efficient

@aakashguptaOct 13

Meta gave its AI away for free for years. This week they finally built one good enough to charge for, and they won't sell it. There's no price. You literally cannot buy Meta's best model at any cost. Sounds backwards until you see what they're doing. Every other AI lab makes money the same way. You pay per word the model reads and writes. OpenAI charges for tokens, Anthropic charges for tokens, Google charges for tokens. The price sheet is the whole business. Meta just opted out of that business. Their smaller new model, Muse Wonder, you can buy like normal, $4 in and $20 out per million tokens, basically Claude pricing. The bigger one, Muse Wonder Full, the one that just tied Claude on coding benchmarks and sits 2 points behind Opus 5.5 overall, has no price sheet at all. The only place it exists is inside Muse Code. And honestly, if you're Zuck, why would you sell it? Meta already has what every other lab is burning billions to get. Distribution. Muse did 2.5 million downloads in 13 days and hit #1 on the App Store. Amazon got nervous enough to block it. WhatsApp puts this thing one tap away from 3 billion people. Once you own the pipes, selling tokens leaks your advantage. Anyone can buy your API output and use it to train their own model, and your frontier intelligence ends up as a line item inside a competitor's product. Keep it locked inside your own apps and the only way anyone touches Meta's best AI is by becoming a Meta user. Meta spent two years lecturing the industry about open AI, Llama weights free for anyone to download. The lectures stopped the same month their models got good.

@rileybrownOct 13

Okay after further testing Muse Wonder might actually be better than Claude Sonnet 3.7

@aakashguptaOct 13

Mark Zuckerberg just said his used computer chips are worth more than what he paid for them. Think about how insane that sounds. Nothing with a power cord works that way. Your car loses value, your phone loses value, your laptop is basically worthless in four years. Zuck is saying the chips running Meta's data centers broke that rule. And this isn't a throwaway line. It's a direct shot at the biggest bear case in the market. Michael Burry's whole argument against the AI giants is that they're pretending their chips last longer than they do so the earnings look better. He pegged the overstatement at $176 billion through 2028. Pretty much every AI bubble take eventually comes back to that same premise, the chips rot fast and the accounting hides it. Zuck went on camera and argued the opposite. Meta keeps running GPUs long after the accountants have written them down to zero, and because the whole world is starving for compute, those "worthless" chips are still earning every single day. He says that's exactly why Meta builds its own data centers instead of renting from the cloud. He's also betting like he believes it. Wang put the cost of running one gigawatt cluster at around $15 billion a year, with $50-60 billion just to build it. Meta is scaling past five gigawatts. Zuckerberg committed somewhere between $100 and $200 billion before Muse Wonder had proven a thing, and when John asked about the downside, his answer was basically that for this to fail, all of AI has to fail. The thing is, Burry and Zuckerberg can both be right for a while. When compute is scarce, old chips hold value and stretching their useful life on the books is defensible. But the day supply catches up with demand, every one of those extended depreciation schedules snaps back at once, across every big tech balance sheet at the same time. So this whole fight comes down to timing. Zuck thinks the shortage lasts years. Burry thinks it ends in quarters, and whoever's wrong is wrong by hundreds of billions of dollars.

@theoOct 13

Okay I’m finally gonna go film the Muse Wonder video. What questions do you guys have?

@rileybrownOct 13

Used Muse Wonder for 5 hours. Was gonna make a video. No need, back to codex and claude code.

@theoOct 12

Having the "Full" version of Muse Wonder restricted to sub-only with no price info available kinda sucks

@kplikethebirdOct 12

Preliminary spot check on the writing quality of Muse Wonder for @every: readability is clocking in at a respectable 8th grade, compared with 7 .76 for Astra and 7.5 for Fable 5.1. Reading level != writing quality, but it is an interesting metric to track while, for now at least, we interact with these models primarily via text 👀