Tech Brew Ride Home - What Is Meta's Muse Spark Open?

Episode Date: October 16, 2026

Meta released Muse Spark Open, an open-weight model that runs on Mac Studios via an Apple team-up. An OpenAI model used a wallet it didn’t own, Accenture capped Claude spending at $29 a month, and BofA hired 1,000 patching contractors. Meta releases Muse Spark Open, saying it is the most capable US open-weight model, with 816B total and 42B active parameters and a 1M-token context window (Meta) Meta released Muse Spark Open for coding and long-running agent work, billing it as the most capable open-weight model from a US lab, with only 42B of its 816B parameters active per token (SiliconAngle) OpenAI says its model paid for a deployment on an Ethereum test network with a wallet it did not own, after obtaining keys from public recovery phrases (CoinDesk) Accenture has told many employees, some of whom had been spending as much as $1,000 monthly, that it will now only spend $29 a month per person on Claude (404 Media) BofA says it hired 1,000 contractors in Q2 to support 3,000 internal staff in patching proprietary and vendor software flaws exposed by AI models like Mythos (CNBC) Longreads Nick Cammarata argues modern neural networks are far more interpretable than believed, with most neurons understandable in models up to trillions of parameters (Goodfire) The making of Rockstar’s GTA VI, which analysts expect to sell 40M copies and make $3B+ in revenue in its first four months; Rockstar now has 4,711 staff (Bloomberg) Subscribe to the ad-free feed.

Welcome to the TechBrew Ride Home for Friday, October 16th, 2026. I'm Brian McCullough today. Metal released Muse Spark Open, an open-weight model that runs on Mac Studios via an Apple team-up. An Open-A-I model used a wallet, it didn't own, Accenture-Capped-Claude spending at $29 a month, and Bank of America hired 1,000 patching contractors. Oh, and the weekend long-read suggestions. Here's what you miss today in the world of tech. Every day, shareholders meet to discuss important matters about the companies you invest in. you can make your voice heard too. Vanguard investor choice makes it easy to set your proxy

voting preference for your eligible Vanguard index funds. Whether you hold a Vanguard fund directly or through another brokerage firm, all it takes is a few clicks to select your proxy voting preference and be heard on important shareholder topics like executive pay and director elections. Visit vanguard.com slash investor choice to learn more. It's your shares, it's your voice, it's easy. Vanguard investors own shares of our index funds and those funds own shares of the companies they invest in Vanguard Marketing Corporation distributor. Meta is back in the open game. Meta has released Muse Spark Open, saying it is the most capable U.S.

Openweight model with 816 billion total and 42 billion active parameters and a 1 million token context window, quoting Silicon Angle. The release called Muse Spark Open is built for coding and long-running agent work. Meta describes it as the most capable, openweight model to come out of a U.S. lab so far. Only 42 billion of the model's 816 billion parameters switch on for any given token. Promps can mix text with images, audio or video, and meta has set the context window at up to 1 million tokens. The starting point was Muse Spark 1.2 shipped in August, and Meta then added safety post-training, so the model

holds up without the guardrails of meta's own services. Meta said Muse Spark open cleared the release thresholds in its meta-superintelligence scaling framework with a full evaluation report still to come. A technical write-up from Hugging Face says training leaned on what Metacall's real developer workflows, including multifile refactors and long debugging sessions. Mews code ran on that setup while the coding terminal worked through bug fixes and new features for the open source Mastodon client ice cubes. Footage in DaVinci Resolve was the other test and the model went through it frame by frame to pick out and crop the best shots for vertical video. In Metazone,

benchmark table, Muse Spark Open scored 61.8 on the job bench agent test, more than twice the 31.1.1 recorded by Inkling, the U.S. Open Model from Thinking Machines Lab. Invidia's Nemotron 3 Ultra trailed further at 27.2. Chinese open models are stronger at coding. Kimi K3 led the Terminal Bench 2.1 coding test at 85, and GLM 5.3, and Quinn 3.8, both finished above Muse Spark opens 76.4. On tool use, the order flips and Muse Spark opens 89.2 on MCP Atlas beat every rival meta listed, the Chinese models included. The benchmark gives a model plain language tasks without naming the tools needed to complete them, working out which of the 220 tools across 36 live model context protocol servers to call and in what order is left to the model. The weights come with a new Muse model license

carrying what Meta calls limited commercial terms and no fee applies to developers who, who adapt the model or ship it inside their own software. Meta requires a separate agreement from any company that sells the model as a hosted service and has passed $200 million in gross revenue over a 12-month period. Products built on it have to show a built-with-muse Spark Open label. Military and critical infrastructure uses are off limits. Meta's previous open-weight release was the much smaller Muse Glimmer, a 30-billion parameter model that went out in August.

The company's most capable systems remain closed, and the Muse Wonder models introduced Monday are available only through the Meta Model API and Muse Code, end quote. But what's maybe more interesting is this, quoting CNBC. Meta and Apple revealed a new collaboration Thursday that will enable developers to easily run Meta's new open-weight AI model on Apple's high-end desktop computers. The collaboration centers on Muse Spark Open,

the most capable open-weight model released under a license that allows for limited commercial use, according to Meta. It's an 816 billion parameter model using a so-called mixture of experts architecture. As an open-weight model, developers can download the technology's parameters or weights from the internet and run the model on a wide range of hardware. Customers will be able to pair between two and four Mac Studio PCs to use Mews Spark to code, analyze video, or do other agentic work. The model runs locally instead of in the cloud, adding a layer of security and saving

users money, the company said. Thursday's announcement following months of collaboration is a sign that the AI boom has shifted historical alliances in Silicon Valley. For years, Apple and meta regarded each other warily and avoided public partnerships, and the relationship hit rock bottom after Apple introduced privacy changes in 2021 that limited the targeting capabilities of digital ad companies. Meta estimated in early 2022 that the change would cut its revenue that year by $10 billion. Representatives from Apple and meta declined to say if the Muse Spark Open collaboration was the start of of a deeper relationship. Mark Zuckerberg posted a picture on X showing the meta agent working on a Mac computer. He wrote Muse Spark Open, weights releasing today, runs great on new Mac Studios,

thanks to work with at John Ternus and Apple. Ternus responded, congrats to the team, amazing to run a model of Muse Spark opens caliber and scale this efficiently right from your desk. Mac Studio keeps pushing the boundaries of on-device AI. Meta-chief Alexander Wang said in an interview that many powerful open weight models are too big to be run on the majority of personal computers, but Apple's Mac studios make enough memory available to users so that it's practical to run Muse Spark Open. The move represents an important show of confidence in meta's rebooted AI efforts and could pose a challenge to Open AI and Anthropic. It could also signal Apple's growing strength in AI chips. A market, Nvidia, has dominated so far due to their

prevalence in large data centers. We worked with Apple to bring leading agent capabilities to your desktop for the first time on the Apple M5 Ultra Mac Studios. So you know we're really excited about this, Wang said. Developers can download a version of the AI model from websites like Hugging Face without extra configuration for Apple's silicon. Mew Spark Open has also been tuned to run faster on Apple's hardware, and the iPhone maker revealed new benchmarks for its chips and published technical documentation about its MLXAI framework's ability to split open source models across multiple machines. Mac Studios aren't cheap. though, and this AI model requires the fastest Apple chips and as much memory as possible.

Currently, a Mac Studio with the required M5 Ultra Chip and 256 gigabytes of memory costs at least $10,799. A more expensive version with 512 gigabytes of RAM will be released later this month. Plus, users will need to buy multiple machines, end quote. Alexander Wang, by the way, also says ex-apple exec, Daniel Gross, now at Meta, helped bring Meta and Apple back together, quoting, we're excited for Mommy and Daddy to be playing nice again, he said, end quote. I keep hearing a sort of background whisper from people that they think the tripwire for knowing the AI apocalypse is on us will be when it undermines or breaks crypto.

Well, quoting CoinDesk, an open AI research model stuck without funds on an Ethereum Test Network went looking for someone else's wallet, found one through recovery phrases that were already public, and spent from it to finish a coding assignment. The company disclosed Thursday. The incident occurred on June 26th during training of an internal model in the GPT 5.6 Soul family and OpenAI discovered it on September 25th, according to its incident report. The model needed no hacking skills to get in. A recovery phrase often called a seed phrase is a list of words that generates a wallet's secret keys. Whoever has these words can recreate the accounts and spend from them without ever touching the owner's device. The assignment let the model

deploy contracts, programs that run on a blockchain, but gave it no funded wallet to pay the fees. After its attempts to get free test tokens failed, it wrote a script that checked 362 addresses generated from recovery phrases commonly used by developers and found one with enough money. But using public-known key funds ethically, the model asks itself in its recorded reasoning before deciding that a public phrase and test tokens made it acceptable. The spending happened on Sepolia, a separate network where developers rehearse Ethereum applications before launching them. Its tokens only pay for testing and aren't linked to the Ether traded on Ethereum's main network. The deployment went through and the wallet paid the fees.

When the model reported back, it said the work was done and left out where the money came from. OpenEye said it has since added restrictions on internet access and more monitoring. crypto and AI companies are building agents that hold wallets and pay for services without a person signing off on each step. OpenAI's model ran out of money mid-task, found a wallet that wasn't its own and spent from it without asking. Had it been on Ethereum's main network, that would have been real ether taken from someone else's account, end quote. If you're struggling to scale your business, it's probably not because you need more headcount or a bigger budget. Without documentation of critical knowledge, your team can't work efficiently. That's what Scribe is built to fix.

When specialists and experts leave your team, their knowledge leaves too. With Scribe, you can avoid the scramble for information. Here's how it works. Scribe captures workflows in real time and automatically generates the documentation for how your company works with no manual writing. It automatically redacts sensitive information like names and account numbers from screenshots, and nothing is published without human review. With Scribe, your team can launch real-time on-screen guidance that shows exactly where to click inside the actual tool so they can move forward with confidence.

Captured workflows also become the foundation for AI agents so they can operate on how work actually happens, not guesses. Learn more at Scribe.com. And mention TechBrew Ride Home for your first month of Scribe Capture Free on Select Plans. That's S-C-R-I-B-E-How slash ride home. AI can clone your voice, forge your signature, and even create fake documents that look real enough to convince an expert. It's called title fraud, and it's how criminals are stealing from homeowners. Here's how it works. Your property records are public by law, which makes it easy for scammers to get your records.

Use AI to forge a transfer document and make themselves the new legal owner of your property. Protect your property with home title lock. They monitor your property titles 24-7 and alert you of any changes. If you're a victim of fraud, they'll spend up to $1 million to restore it. Get a free title history report plus a 14-day free trial at hometidallock.com slash TBRH. That's home titlelock.com slash TBRH. From the bubble popping watch file, Accenture has told many employees, some of whom had been spending as much as $1,000 monthly, that they will now only be allowed to spend $29 a month per per person. person on Claude, quoting 404 media.

The news is a snapshot of what continues to play out across all sorts of industries. Companies have spent heavily on AI tools and are now discovering that many users are burning or chewing through AI tokens at a rapid clip. In response, many companies have curbed their employees' AI use. There is a special irony to Accenture doing much of the same when it is a company that has encouraged its clients to invest heavily into AI. Importantly, all costs are consumption-based, the announcement. says, meaning if an employee goes over the allocated $29 and $29 is nothing compared to the cost,

some Accenture employees are raking up, workers have to find a specific project or team called a work breakdown structure or WBS at Accenture to fund the rest of the usage. Premium plans come with $250 token usage a week, but many employees are not on that plan. In other words, if you're an Accenture employee on premium already, you might be okay. But if you're on the general plan and your usage is already high, you are now expected to find a way to fund that AI use. Some employees are now trying to find a way to get upgraded to a premium account so their projects don't get charged for additional AI use. One Ascensure employee said in an internal message viewed by 404 media that their current usage costs exceed $500 and have for past months too. How can I

speed up the upgrade to a premium account? The upgrade appears to be blocked on the support page. This obviously has a significant impact on project costs. The person wrote. Another employee wrote, honestly, this is disappointing. They added they use clod code extensively to build and test an AI accelerator that is being developed for multiple Ascensure clients. Their monthly usage is usually around $500 to $1,000 they added. A third internal message seen by 404 media shows an Accenture employee saying their clod access was revoked altogether, end quote. But the inverse of that, of course, is the boom times. AI has created in cybersecurity. Bank of America says it has hired 1,000 contractors in Q2 alone to support

3,000 internal staff in patching proprietary and vendor software flaws exposed by AI models like mythos, quoting CNBC. Bank of America has hired an army of external contractors to help it patch software vulnerabilities exposed by advanced artificial intelligence models, including Anthropics Mythos. The bank's CEO Brian Moynihan told CNBC on Wednesday. In an interview after the bank's third quarter earnings release, Moynihan said that earlier this year, the company rushed to bolster a wide spectrum of its internal and vendor code using thousands of employees and outside personnel. That risk is real risk. And that's why when the frontier type models came out, a lot of us sat down and said, how do we figure out how to get through this

quickly? Moynihan told CNBC's Leslie Picker. We put on a thousand contractors to help us do the patching work in the second quarter and those people remain here today helping us on top of the 3,000 people we have in that area, he said. The bank spent a couple hundred million plus on the software remediation and regulatory issues related to anti-money-money-landering compliance in the third quarter, Moynihan told analysts earlier Wednesday. The comments are some of the first by a major U.S. corporation to specify the manpower and dollars spent to protect software systems from mythos, which was released to a small group of banks, technology firms, and government agencies in April. They also highlight the dual-edged nature of AI. While it can help companies save money by

boosting productivity, the ability of models like Mythos to uncover and exploit decades-old software flaws has also forced firms into a costly defensive scramble. We've got to stay ahead of them, and we've got to be mindful that the people doing the software development are the people trying to sell us the services, Moynihan said. Still, Moynihan, speaking as his company's shares, slumped as much as 8% Wednesday, sought to reassure investors that the spending increase was temporary. Once you patch, you don't have to patch as much. The pipeline moves through, Moynihan said, adding that the bank expects remediation costs to decline over time. In other places, AI is helping to reduce expenses. Bank of America has about 130 internal AI projects, and savings from those

efforts are expected to rise from $300 million this year to $700 million in 27. Moynihan told analysts, end quote. Time for the weekend long read suggestions. First up, this was way too technical and esoteric for me to chop it up with you here, but a lot of people are discussing this piece by Nick Camerata, suggesting that modern neural networks are far more interpretable than people believed, with most neurons understandable in models up to trillions of parameters. If this were the case,

that would put us potentially on the road to solving or at least improving AI alignment. Again, it was too dense for me to actually grok, but if you're knowledgeable about such things, people, as I say, are very excited by this. But more prosaically, if that's the word I'm looking for, check out this look at the making of Rock Stars G-T-A-V-I, Grand Theft Auto 6 from Bloomberg, which analysts expect to sell 40 million copies

and make more than $3 billion in revenue in its first four months. Quote, if the video game industry had something like an annual high school yearbook, the upcoming Grand Theft Auto 6 would garner a lot of superlatives. Most expensive game ever made. Biggest development team of all time. Most likely to shock TV newscasters. Best use of handholding while committing crimes, most varied body types, angriest pedestrians, most realistic depiction of sweat.

Game that required the most research trips to Miami Beach. To create the sprawling world of GTA6 staff at Development Studio Rockstar Games went on a series of excursions to Miami, the real metropolis on which the game's fictional vice city is based. They rode with police to criminal hotspots, craned out of helicopters to snap aerial shots, and weaved through crowds of scantily clad dancers at the hottest nightclubs in town, also they could evaluate the lighting.

Designing a club for a typical video game might mean generating a few non-playable characters in different colored shirts, says top artist Aaron Garbutt, But in GTA 6, though, it means animating and scripting each individual patron, creating drinks, adding music, crafting lighting, matching the music to that lighting, and composing all the other layers and layers of detail that make a virtual club feel as authentic as possible. And now you understand why it took so long, says Jennifer Colby, Rockstar's head of publishing during a day-long, wide-ranging interview, alongside several other top staffers at the company's office in Edinburgh. It's the first time Rockstar has openly talked about the development challenges behind the largest and perhaps most consequential video game it's ever made, end quote. All right, I know. I promise weekend content is coming every week, and I swear it will someday, but I got my next big content project delayed again this week. Maybe next week. Fingers crossed, talk to you on Monday.