Meta Just Dropped an 816 Billion Parameter AI Anyone Can Download and It Only Needs 2 Mac Studios to Run

-

TL;DR: Meta just open-sourced a 816 billion parameter AI model called Muse Spark Open. It uses a Mixture of Experts architecture so only 42 billion parameters activate per request, and you can run the whole thing locally on two Apple M5 Ultra Mac Studios. Commercial use has restrictions but the weights are free to download today.

I had to re-read this headline three times before I believed it. Meta dropped an 816B parameter model and just let anyone download it.

For context: most of the biggest open models out there cap out around 70 billion parameters. GPT-4 is estimated somewhere north of a trillion but is locked behind an API. And here is Meta putting something in the same weight class onto the internet for free.

What Is Muse Spark Open

Meta calls it Muse Spark Open. It is a Mixture of Experts model, which means not all 816 billion parameters run at once. When you send a prompt, the model activates a routing layer that picks which specialist sub-networks to use. For Muse Spark Open that comes out to about 42 billion active parameters per request. That is still a big model but it is not the absolutely crushing compute requirement the headline suggests.

MoE has been the architecture behind some of the best recent releases including Mistral's Mixtral and reportedly GPT-4 itself. Meta took that playbook and scaled it way up.

The Hardware Story

Meta says the reference setup for running the full model is two Apple M5 Ultra Mac Studios connected via Thunderbolt. That is roughly 384 GB of unified memory between the two machines. The model weights at 4-bit quantization land around 400 GB so you are right at the limit, which is tight but apparently workable.

If you want to run it at full precision (16-bit) you are looking at a proper GPU cluster. But the quantized version on two Mac Studios is a real use case that a small studio or indie developer could actually pull off. A pair of M5 Ultra Mac Studios is not cheap but it is not hyperscaler territory either.

What It Is Good At

Meta is positioning Muse Spark as a general reasoning and generation model. Early benchmarks from the community are putting it close to GPT-4o on most standard evals. Coding, math, and long document summarization all look competitive. Creative writing quality is getting a lot of positive comments in the early test threads I have been following.

The model supports a 128K context window, which gives it serious reach for document analysis tasks. That is on par with what you get from the frontier API models.

The License Details Matter

This is where it gets a little complicated. The weights are free to download and you can run the model locally as much as you want. The commercial restrictions kick in at scale: if your product or service reaches more than 700 million monthly active users you need to go back to Meta for a special commercial license. For basically every startup or independent developer that threshold is not a concern.

There is also a clause prohibiting use in certain regulated industries without additional agreements. It is similar to what Llama models have had in previous versions. Not truly permissive open source, but dramatically more open than anything else at this parameter count.

Why Meta Keeps Doing This

Meta has been the most aggressive big tech company in open sourcing AI. Every Llama release has pushed the open source baseline higher. This is partly strategic: a world where everyone builds on Meta infrastructure and model weights means more engagement, more data feedback loops, more talented researchers who know Meta tools.

But it also puts real competitive pressure on the closed model providers. Hard to charge premium API prices when a comparable model is free to run locally.

How to Get It

The weights are on Hugging Face under the meta-ai organization. The model card has the quantization instructions and the recommended hardware configs. The community has already started building Ollama and llama.cpp integrations so you will not need to write much infrastructure yourself.

What's your experience? Drop a comment below! 👇

Have you tried running any of Meta's open models locally? What hardware are you running them on and what are you using them for?

#MetaAI #MuseSpark #OpenSourceAI #AI #LLM #ArtificialIntelligence #MacStudio #TechNews #AIModel #MixtureOfExperts

Comments

Post a Comment

Popular posts from this blog

Plank You App Support and Privacy

-
Plank You is a simple, ad-free interval timer for planks and core workouts. Set your work and rest times, hold your plank, and let the countdown beeps do the counting. Everything runs on your device. no ads, no accounts, no tracking. How to use Start / Pause / Resume: tap the main button to run or pause the timer. Reset returns to Rep 1 and the start of your work interval. Rep counter — counts up automatically after each full work + rest cycle. Beeps — the last 5 seconds of every phase beep down (5-4-3-2-1), with a longer tone when the phase changes. Changing intervals Tap the settings icon (top right) to choose an interval: 45 / 15 default (45s work, 15s rest) 30 / 30 50 / 10 Custom — set any work and rest time in 5-second steps You can also toggle Beep sounds and Haptics in Settings. FAQ I don't hear any sound. Sound plays even when your ringer is on silent, but make sure the media volume is up and that "Beep sounds" is turned on in Settings. If another app is ...
Read more

WWDC 2026 Preview: Apple Needs to Fix Siri or It's Game Over for Apple Intelligence

-
Apple's WWDC 2026 kicks off on June 8, and the stakes couldn't be higher. While Google announced 900 million Gemini users at I/O last week, Apple is heading into its biggest developer conference with one uncomfortable question hanging over the whole event: what happened to Siri? I've been using Apple devices daily since the original iPhone, and I'll say it plainly — Apple Intelligence has been the most underwhelming software launch in Apple's recent history. WWDC 2026 is Apple's chance to fix that. Here's what they need to announce, what we're expecting, and what a failure would mean. The Problem Apple Needs to Solve When Apple announced Apple Intelligence at WWDC 2024, the demos were impressive. When the features started rolling out through 2025, the reality was a different story. The "new Siri" was slower, less reliable, and often less capable than the old one for basic tasks. The "Personal Context" features that were supposed ...
Read more

This AI Startup Is Worth $26 Billion and Writes 90% of Its Own Code — Should Software Engineers Be Worried?

-
I've been watching the AI coding space very closely for the past year, and every few months something happens that forces me to recalibrate my sense of how fast things are moving. This week, that moment came from Cognition AI — the startup behind Devin, the AI software engineer — which just announced a $1 billion funding round at a $26 billion valuation . But the number that actually stopped me cold wasn't the valuation. It was this: 90% of Cognition's own code is now written by Devin. From $10B to $26B in Eight Months — The Numbers Are Staggering Let's put this in perspective. Just eight months ago, in September 2025, Cognition closed a $400 million round at a $10.2 billion valuation. Today, that valuation has more than doubled to $26 billion. The new round was co-led by Lux Capital, General Catalyst, and 8VC, with Ribbit Capital, Atreides Management, and Peter Thiel's Founders Fund also participating. But what's really driving this isn't hype — it...
Read more