😺 Did Zuck reboot the race?

 Sign Up · Advertise

Welcome, humans.

OpenAI’s Codex just hit 3M weekly users, and Sam Altman celebrated the milestone the modern way: by smashing the reset rate limits button. 

He said OpenAI plans to do that every time Codex gains another 1M users until it reaches 10M. In Codex-land, that means the free breadsticks are back.

The joke is that these resets have basically become part of the Codex experience that has routinely become more like a national holiday among users.

And yes, the people’s champion here is Thibault “Tibo” Sottiaux from the Codex team, who’s become closely tied to these regular resets and the whole “keep the coders fed” vibe. (Many thanks from me, as well. PrEsS tHe BuTtOn! -CN)

It’s a funny little meme, but it also says something real: Right now, developers just want agent that stays usable when everyone shows up at once.

Here’s what happened in AI today:

• The CIA reportedly used Ghost Murmur to find a downed U.S. airman in Iran.

• Z.ai released GLM-5.1, an open-source model built for longer autonomous work.

• Perplexity’s ARR reportedly topped $450 million.

• X rolled out Grok-powered translation and photo editing.

• Gmail users have a careful decision to make.

…and a whole lot more you can read about in our Midweek Around the Horn digest!

P.S: Want to reach 675,000 AI-hungry readers? Click here to advertise with us.

P.P.S: Love robots? We’re starting a new robotics newsletter! Sign up early here.

😺 Meta just rebooted its AI race with Muse Spark

Meta is officially back in the AI ring. The company just launched Muse Spark, the first major model from its new Meta Superintelligence Labs, and its first big AI release since Mark Zuckerberg reshuffled the org and brought in Alexandr Wang to help get things back on track.

The pitch is pretty straightforward: Muse Spark now powers the Meta AI app and Meta.ai, with rollout coming to WhatsApp, Instagram, Facebook, Messenger, and Meta’s AI glasses in the coming weeks. Meta says it’s multimodal, built for speed, and offers both an Instant mode for quick answers and a Thinking mode for harder questions.

That’s important because Meta isn’t trying to win the AI race the same way OpenAI and Anthropic are. It’s trying to shove AI into every app you already open 40 times a day. Your messages, your feed, your glasses, and probably eventually your very questionable vacation planning.

And this model is no slouch, scoring a 53 to debut at 4 on the Artificial Analysis Intelligence Index behind only Gemini 3.1 Pro, GPT-5.4, and Claude Opus 4.6. That’s a powerful comeback from a company whose most recent model, Llama 4 Maverick, scored only an 18. 

Meta finding its way back into serious AI conversation is no small feat. Countless people in the AI space had all but written them off. And to be clear, this didn’t tap the top spot, and that’s still a hefty climb for Zuck & Co. 

But it’s a whole lot closer than it was yesterday.

There are a few catches, though:

• Meta may open-source some future models, while keeping its strongest systems proprietary.

• Public access still looks limited for now, with the biggest push happening through Meta’s products rather than a wide-open standalone release.

• And in a funny little sign that things are getting tense internally, journalist Jyoti Mann reported that Meta shut down its internal AI token leaderboard, dubbed  Claudeonomics, after data from the dashboard leaked externally. Tough day for the office scoreboard.

Why this matters: Meta may not need the best model if it has the best distribution. If Muse Spark is good enough and lands inside apps used by billions, that could matter more than winning benchmark bragging rights for a week.

Our take: This feels less like a knockout punch and more like Meta proving it can still throw one. But the bigger story is the strategy shift: Meta seems to be moving from “open for everyone” to “open where useful, closed where it counts.” Smart? Maybe. A pretty major philosophical turn? Definitely. Either way, Corey is not-so-patiently awaiting its arrival in his Meta Ray-Ban smartglasses, and we’re excited to see what else they’re cooking up.

The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.

This guide distills 10 AI strategies from industry leaders that are transforming marketing.

• Learn how HubSpot's engineering team achieved 15-20% productivity gains with AI

• Learn how AI-driven emails achieved 94% higher conversion rates

• Discover 7 ways to enhance your marketing strategy with AI.

Outpace your competitors by mastering AI.

Get Your Free Guide

🎓 AI Skill of the Day: Break Big Tasks Into Stages

One of the easiest ways to get mediocre results from AI is to hand it a huge job and expect one great answer. That’s usually how you get something polished, generic, and not all that useful. 

AI works better when you give it one clear step at a time. Instead of asking for the finished product right away, break the work into pieces. 

If you’re writing, start with the outline. Then pressure-test the outline. Then draft one section. Then refine the tone for the audience. Then pull out action items, risks, or next steps. That works for writing, research, presentations, hiring plans, basically anything with more than one moving part.

It sounds simple, but it’s one of the biggest unlocks in using AI well. You get better output, more control, and way fewer responses that look right without actually being right. 

The people getting the most out of AI usually aren’t asking smarter one-shot prompts, they’re just better at turning messy work into smaller steps.

Total AI beginner? Start here (goes with this video).

Have a specific skill you want to learn? Request it here. 

NEW from The Neuron: Why Reflection AI Is Suddenly Everywhere

Cue the Jaws music. So Meta just went semi-closed-source (see above). Which raises a pretty obvious question: if the company that championed open AI is locking the doors, who's picking up the torch? Well, meet the guy who co-created AlphaGo at DeepMind, raised $2B from NVIDIA, and is now building frontier open-weight models at Reflection AI specifically to be America's answer to DeepSeek.

In our latest podcast episode, Ioannis Antonoglou explains why every frontier open model right now comes from China, how mixture-of-experts lets trillion-parameter models actually run, and why he believes open source will eventually surpass the closed labs—even the one that just abandoned ship.

Watch and/or listen: YouTube | Spotify | Apple Podcasts

FROM OUR PARTNERS

Editors Pick - Slack

Why We Love It: Slackbot is your always-on personal agent for work — and it's already built right into Slack. Discover Slackbot

🍪 Treats to Try

1. Anthropic launched Claude Managed Agents, a suite of APIs for building cloud-hosted AI agents without infrastructure setup (price: token rates plus $0.08 per-session-hour); two YouTube creators explain how to use it, with split reactions: Nick Saraev (371K subs) said it kills n8n 👀; Nate Herk (639K subs) called it disappointing for power users since it lacks scheduled triggers and still needs external glue for always-on automation.

2. Lessie helps you find the right creators, leads, investors, and hires by typing who you want in plain English. $39-$299/mo.

3. SnapRewrite lets you highlight messy text anywhere on your Mac and instantly swap in a better rewrite. Free plan to $5.99 lifetime plan.

4. VibeSonic turns your voice into notes, tasks, and transcriptions anywhere your cursor is on your Mac, so you can speak instead of typing and keep moving through your workflow. Free to try, then Pro is paid only rn ($19.95 lifetime for 1 device).

5. Career-Ops helps you run your job search like a system by scanning roles, generating tailored CVs and PDFs, and managing the whole process from one dashboard. No pricing details.

📰 Around the Horn

• The CIA reportedly used a secret AI tool called Ghost Murmur to help locate and rescue a downed U.S. airman in Iran.

• Z.ai released GLM-5.1, an open-source model it said could work on a single task for up to eight hours and outperform top closed models on some coding benchmarks. 

• Perplexity’s annual recurring revenue reportedly topped $450 million after a pricing shift and a push into AI agents.

• X rolled out automatic post translation and AI photo editing powered by Grok.

• Gmail users have to make a careful decision as Google pushed major Gemini-powered upgrades to Gmail.

FROM OUR PARTNERS

What If Your AI-Built Product Actually Made You Money Every Month?

Not a side project collecting dust. Not a demo you show at meetups. A real product with real users who pay recurring revenue — built with AI, structured by CTO judgment, and 100% owned by you. The AI Founder Accelerator makes that a system, not a dream.

→ Request Early Access

🧩 Thursday Trivia

One is AI, and one is real. Which is Which? Vote Below!

A.

B.

A Cat’s Commentary

Trivia answer: B is AI, and A is real…

P.S: Before you go… have you subscribed to our YouTube Channel? If not, can you?