- AI Fire
- Posts
- ⚡ Meta’s Biggest Open AI Model Yet
Sponsored by
Plus: AI Personal Agent Guide: Build Your Own or Use These Ready-to-Deploy Ones (Prompts + Real Demos)
An 816B-parameter AI model running on Mac Studios!? Meta just dropped Muse Spark Open for coding and AI agents. And developers can download its weights.
What's on FIRE 🔥
IN PARTNERSHIP WITH LEADPOST
Ad costs climb every quarter, and the same visitors keep slipping away unnamed. Meanwhile the cheapest leads you'll get all month are already on your site. You just can't see them.
LeadPost turns anonymous sessions into contact records you own: name, email, postal address, company, job title. It runs on the traffic you're already buying and pushes leads into your CRM automatically, so email, ads, and direct mail can pick up from there. One customer traced $270,000 in sales back to a $4,000 spend. Free for 14 days for AI Fire readers, with no credit card. You'll see real leads out of your own traffic before you decide anything.
14-day free trial, up to 200 leads. No credit card required. U.S. website visitors only. Works best on sites with at least 1,000 unique monthly visitors
AI INSIGHTS
Meta released Muse Spark Open, which it calls the most capable U.S.-developed open-weight AI model to date. It is built for coding, long-running agents, and complex software work. Main points:
816B total parameters, but only 5.1% active per token using a sparse Mixture-of-Experts architecture.
1M-token context window for huge codebases and long agent workflows.
Natively understands text, images, audio, and video.
Can be self-hosted, so prompts and responses can stay inside your own infrastructure.
Meta says it leads several major open-model benchmarks, including JobBench, MCP-Atlas, HLE-Diamond, ZeroBench, and SimpleVQA, although it does not win every coding benchmark.
The wild part is the Apple setup. Meta worked with Apple to optimize Muse Spark Open for M5 Ultra Mac Studios. Two Mac Studios with 512 GB of unified memory each can operate together with roughly 1 TB of unified memory.
It is available now through Hugging Face and cloud/API providers including Groq, OpenRouter, Together AI, Fireworks, Baseten, Modal, and Nebius.
🎁 Today's Trivia - Vote, Learn & Win!
Get a 3-month membership at AI Fire Academy (500+ AI Workflows, AI Tutorials, AI Case Studies) just by answering the poll.
What makes Spark Open’s 816B parameters unusual? - |
PRESENTED BY WISPR FLOW
Leave Granola and get up to 12 months free of Wispr Flow Notetaker + Dictation
If you have paid time left on an individual Granola plan, we'll match it with a Wispr Flow subscription that includes Notetaker and dictation, and add bonus time, up to 12 months total. Sign in or create a Wispr account and submit proof of your plan to check eligibility.
AI PERSONAL AGENT GUIDE
Today, I'll show you 2 ways to build your own AI Personal Assistant:
Ready-made AI agents: Meta Muse, ChatGPT Dots, Grok Bot. Easier to set up, with built-in memory and proactive features.
DIY AI agents: Claude Code, Codex, ChatGPT Work. More control over your data, workflows, and customization.
These agents could remember everything and actually help you before you even ask. So which one should you use? And how do you make them actually useful in your daily life?
In this guide, I'll walk you through both approaches, share real prompts you can use today, and show you 2 practical demos so you can see what these AI assistants can actually do.
TODAY IN AI
AI HIGHLIGHTS
🌐 This prompt creates a 15-second 3D flyover of your website. If normal screen recordings feel boring to you, this is a much more eye-catching way to present it.
🧠 Greg Isenberg just shared 21 AI industry takeaways after 48 hours with execs behind some of tech’s biggest companies. Also yes, Jensen Huang made the selfie.
⚡ Product Hunt’s CEO has a warning for AI builders: every new model can upgrade your product or make it irrelevant overnight. His answer is a skill called “buoyancy.”
🛡️ Scam texts are getting harder to spot, so Google is putting Gemini on patrol inside Messages. It can read unknown-sender chats and judge if something feels off.
🍎 Apple may finally launch AI in China this December, but it’s arriving with two-year-old features and no new Siri AI. So don’t expect Apple’s newest AI experience yet.
🗳️ Anthropic just detailed its 2026 midterm safeguards. Only 0.51% of 250K sampled US Claude chats involved elections or voting, while misuse remained extremely rare.
💰 Big AI Deal: OpenAI and Anthropic are backing a $15 million math initiative from the Simons Foundation, giving researchers funding, AI tools, workshops, and compute as the field changes fast.
NEW EMPOWERED AI TOOLS
🎙️ GPT-Live-1.1 adds 13 new AI voices, better accents, and smoother conversations across ChatGPT Voice, Codex, and dots.
🎨 FLUX 3 Image Klein generates and edits images in native 4K, runs on a single consumer GPU, and lets developers customize the model.
🎨 Figma Specs lets you teach AI agents your brand’s design rules, reuse design libraries, and turn designs into interactive prototypes.
AI BREAKTHROUGH
NVIDIA just released Nemotron 3.5 Super VL, an open multimodal model built for specialized AI agents that need to reason across text, images, documents, video, and tools in one workflow. Main points:
It has 120B total parameters but activates only 12B per token, using a Mixture-of-Experts architecture to keep inference efficient.
It supports up to 1M tokens of context, useful for long documents, video analysis, research, and persistent agent workflows.
NVIDIA says it reaches 600+ output tokens/sec on some agentic workloads and delivers up to 5× higher throughput than comparable open models in its class.
Companies including Scale AI, Ramp, and AI21 have already post-trained it for specialized agents, with NVIDIA reporting cases of 40× lower inference cost, 10× lower latency, and 40% higher accuracy.
On Plotly’s independent analytics benchmark, it answered a typical task in 12 seconds, the fastest of 28 tested models, while scoring 70%.
Developers can download the weights, fine-tune it with NeMo, run it locally or in the cloud, and customize it around their own data, tools, policies, and workflows.
A CRM so smart, it updates itself.
HubSpot's CRM is so smart, it now updates itself. Calls get logged and summarized the moment they end. New leads get researched and a first outreach drafted, automatically. Deals move forward on their own, with next steps flagged before you have to ask.
Even the one repetitive task you've been meaning to fix can now run itself, no code required. Customers using it are already seeing the difference.
This isn't the CRM you have to work hard to figure out. It's the one already working before you log in.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply