Alexandr Wang

if i was to work for any lab it would be meta. as much as i love claude models, how autonomous grokbot is and the reliability of codex, its clear the world is heading towards agents working in the background as we get to spend more time doing what we love. community building and entertainment will be huge (it already is). creator economy is about to roadster flip over everything. meta distribution via instagram and partnering up with apple means they’re going to own a major % of the consumer end. mark zuckerberg recently said something in the recent cheeky pint podcast that really stuck with me. "the highest use of super intelligence is invention and creation, not automation". the more i use agents the more im realising this myself. more time to connect with loved ones, more time to work on passion projects and more time to solve harder problems. no one cares how much faster you’re flying through excel spreadsheets. or how much money you’re making. its really time to have fun. you need to dream bigger. thats all i’ve ever wanted to do and if progress in si stops here i would still be fully content with life. and the crazy thing is i haven’t even tried muse yet (im from uk/eu and they’re setting us back 100 years right now🥲). but i love how meta trained the model to work for personal use case in mind and this aligns with my philosophy: “more human, by design”. i see where this all heading so much clearer now. personal super intelligence. i dont even like saying its bigger than the internet anymore because thats obvious. truly once in a civilisation moment like alexandr said.

I didn't expect that, but I'm actually kinda really enjoying using Muse. I have some feedback to share, but it's a pleasant model to work with.

Okay my Muse Wonder Full benchmark is now complete with @VulcanBench - busy weekend so might not get to put all the data and model cards together until next week. But I wanted to share a quick peek at the combined score and how it compares to other models. The TL;DR is, with Muse Wonder Full, Medium effort seems to be the sweet spot, but, you need to take tokens-per-task and cost-per-task into account when deciding what model to use. In the two tables below, first is combined accuracy score, second is time per task. Right now Muse Wonder Full is quite a bit slower than Opus 5.5, much slower than GPT-6.1 Sol, and way slower than GPT-6 Astra. Overall, based on what I'm seeing, I would use Opus 5.5 or GPT-6.1 Sol over Muse Full at this point. You'll get the same or better accuracy faster and cheaper. More to come in model cards and deeper analysis on the site, but it's the weekend and I'm hanging with family more than doing benchmarks so might be a day or two. You should do the same! 🖖

> be Muse Spark Open > Meta's Llama 4 disappoints everyone > Meta goes closed with Muse Spark > Zuck writes a $100-200 billion check > Muse Wonder ships on Monday > 3 days later Meta drops you with open weights > 816 billion parameters > only 42 billion active at a time > 1 million token context > built for coding and long agent tasks > runs on two Mac Studios on a desk > Meta calls you the most capable US open-weight model ever > every Chinese lab reading the model card America is back in the open-weight game.

1.4.8 made Muse Code a lot more token-efficient, so we wanted you to feel it right away. 🎉 Everyone's usage is reset, effective now. Have fun.

Congrats to the team! Amazing to run a model of Muse Spark Open's caliber and scale this efficiently, right from your desk. Mac Studio keeps pushing the boundaries of on-device AI!

Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.

OPENAI to EMPLOYEES: If someone is trying to seduce you, first consider — could they be a spy??? OpenAI is warning its employees about honeypot spies and strangers with children in playgrounds ... a very fun one with @georgia_wells and @MeghanBobrowsky: wsj.com/tech/ai/ai-companies-play-cat-and-mouse-with-secret-stealing-spies-160eec6b

My AI agent doesn't live in a chat window. She lives on my studio wall 🌅 Pi 5 + laser projector + No screen, no app — just her, floating on the wall, changing state as she works. We talk out loud — she listens, and answers back in her own voice. @alexandr_wang @natfriedman — your @muse SDK escaped the gadget and moved in! Still a work in progress — new senses coming..

Pros: - writing & "voice" - capabilities are on par with 6.1 sol (in non specialized things like computer use or 3d, they didn't RL on those this time) - orchestration Cons: - seriously wtf is this naming scheme you guys are making gpt-5.2-codex-high-fast seem sane - price/usage limits (tho they're working hard on fixing this) - can only use the good one (full) in muse code through the sub. no API

LIVE: The full How I AI bench on Muse Wonder (and Wonder Full) - the new frontier model by @AIatMeta I really loved how it handled agentic tasks, really hated what it designed. Still TBD: will I reach for it every day? Full ep on YT: youtube.com/watch?v=dChiPsd0uF4

Gemini 4 Argon has proven to be outright unusable for real code work in my codebases. It aggressively hallucinates about how systems work.

We’re exploring how AI can accelerate health science research. While AI has made significant breakthroughs in formal sciences like mathematics, health data science depends on imperfect real-world data and consequential choices about cohorts, variables, and causal assumptions. To explore how AI can assist health researchers, we built a human-led research scaffold powered by Muse Spark 1.3 and Muse Wonder with access to 5M retrospective de-identified electronic health records and publicly available data. A cohort of medical experts used the system to conduct original research, accelerate hypothesis generation, code execution, and manuscript drafting while maintaining full authority over scientific decisions. Today, we’re sharing 10 research manuscripts produced through this exploration, spanning cardiology, oncology, infectious disease, emergency medicine, and more. Read more in our research blog: go.meta.me/3U06wvD

Gemini 4 Argon High vs Muse Wonder Full 🍉, the crazy thing is that Muse Wonder wasn't even trained specifically for 3D, yet it still beats Gemini 4 in most of my tests, with both costing about the same per task, In this article I share a much deeper look at my tests and how Wonder compares with other frontier models like Opus 5.5, GPT-6 Astra and more, A little spoiler, Wonder Full Max and Xhigh are really good, they're definitely expensive but well worth trying!