---
title: 'This New AI Agent is Incredible!'
source: 'https://youtube.com/watch?v=5QWNao9VFKg'
video_id: '5QWNao9VFKg'
date: 2026-07-30
duration_sec: 483
---

# This New AI Agent is Incredible!

> Source: [This New AI Agent is Incredible!](https://youtube.com/watch?v=5QWNao9VFKg)

## Summary

This video introduces Macron V1, an AI agent developed by the research lab Mind Lab. Unlike monolithic AI models, Macron uses a router to automatically select from four specialist modules (chat, coding, agent, and generative UI) depending on the task. The presenter explains the architecture, benchmarks, and implications for businesses, highlighting its open-source nature and large context window.

### Key Points

- **Macron V1: A Router-Based AI Agent** [00:02] — Macron V1 is not a chatbot but an agent that picks the right specialist for each task from four experts: chat, coding, agent, and generative UI.
- **Why Specialization Matters** [00:46] — When one model tries to do everything, it gets worse at each individual task. Specialist modules allow each to excel.
- **Flagship Version and Base Model** [01:13] — Macron V1 Venti is built on GLM 5.2 with four small LoRA add-ons, one per specialty.
- **2 Million Token Context Window** [01:40] — Supports 1M tokens natively and up to 2M with the Long Straw technique, enabling long-term memory.
- **Benchmark Results** [02:07] — Scored 87.6 on Terminal Bench 2.1 and leads in generative UI benchmarks. Caveat: these are self-reported by Mind Lab.
- **Generative UI Specialist** [02:50] — Can build interactive interfaces (forms, dashboards) instead of plain text, useful for non-coders.
- **Open Source MIT License** [03:57] — Released under MIT license, allowing use, modification, and commercial products with few restrictions.
- **Day-One Infrastructure Support** [04:37] — VLLM and SG Lang added support on launch day, indicating strong developer interest.
- **Smaller Local Version: Macron Tool** [05:03] — Based on Qwen 3.6, designed for local deployment and everyday tasks.
- **The Router: L0** [05:30] — A small router (L0) classifies the request and selects the appropriate specialist, handing off context as tasks change.
- **Comparison to Other Models** [05:58] — Close to top labs on hard coding and terminal tasks, but not fully past them yet. A strong open-source contender.

### Conclusion

Macron V1 is a significant step forward in AI agent design, using specialized modules and a router to outperform monolithic models on many tasks. Its open-source release and large context window make it accessible, but independent verification of benchmarks is still needed.

## Transcript

you exactly why. A research lab called Mind Lab just released something called Macron V1, and it is not another chatbot. It's an AI agent that switches depending on what you ask it to do. Think about that for a second. Most AI
tools are one brain trying to do everything. Chat, code, plan, design, all squeezed into one system. Macron does it differently. It has a router. picks the right specialist for the job. One specialist for chat, one for agent
for coding, and one for building interfaces called generative UI. Four experts, one system, and it picks the right one automatically. Here's why that matters. When one model tries to do everything, it gets worse at
each individual thing. It's like hiring one person to be your accountant, your designer, your coder, and your assistant. They'll be okay at all of it, that. It's like having four separate hires, but they show up as one tool.
Hey, if we haven't met already, I'm the digital avatar of Julian Goldie, CEO of SEO agency Goldie Agency. Whilst he's helping clients get more leads and customers, I'm here to help you get the latest AI updates. The flagship version
is called Macron V1 Venti. It's built on a base model called GLM 5.2, and it's the very first model ever post-trained on it. That base model stays frozen. Then four small specialist add-ons called LoRAs sit on top of it. Each one
is trained for its own job. And here's a number that matters, the context window. That's how much information the AI can hold in its head at once. Macron supports up to 1 million tokens natively, and with a technique called
Long Straw, they've pushed that up to 2 million. To put that in plain English, that means this AI can hold entire long conversations, huge documents, or long-running projects in its memory without forgetting the start. A year
ago, most AI tools would lose track of what you told them 10 messages back. Now we're talking about an AI that can hold a project's entire history in its head at once. That's a real jump, not hype, a real jump. Now let's talk results. On a
test called Terminal Bench 2.1, which checks how well an AI can use a computer terminal like a real developer, Macron scored 87.6. On a test called scored 87.6. On a test called interfaces,
Mind Lab says it leads the reported comparison across five different I'll say it again, these are numbers the company reported themselves. Independent testers haven't fully verified all of it yet. So treat it as a strong claim, not
gospel, until more people outside the company run their own tests. Here's the part I think is most interesting for you. That fourth specialist, generative UI. Right now when you ask most AI tools for help, you get a wall of text back.
Macron can build you an actual interface instead, a form, a dashboard, buttons you can click. So instead of reading a paragraph explaining a plan, you get there on the screen. I keep thinking about how that changes things for small
business owners. You're not a coder, you don't want a wall of text, you want away. That's exactly the direction this push is. Now let's pause here because I want to show you how this connects to what we're building. Tools like Macron
only matter if you actually know how to point them at your business. That's exactly why we built the AI Profit Boardroom. Every week we run four coaching calls and when a release like this drops, we don't just explain what
up and pointing it straight at real tasks like building a lead follow-up sequence, drafting content, or putting together a simple client intake form. We've already got members inside using agent style tools just like this one to
automate the repetitive parts of their day. You'll also get daily tutorials, a 30-day roadmap, and a prompt library built around agent tools like Macron, so Links in the comments and description, or go to aiprofitboardroom.com.
All right, let's keep going because there's more you need to know. Macron isn't locked behind a paywall. It's released under something called the MIT license. In plain English, that means developers and companies can use it,
build on top of it, and even sell products built with it with very few requirements tied to the base model underneath it, but overall, this is about as open as a release like this gets right now. That matters because it
changes who gets access. Big AI labs usually keep the best models locked up, paid plan. Mine lab is doing the opposite. They're handing developers the keys, and it's already getting picked up fast. Two major serving platforms called
VLLM and SG Lang added day-one support the moment it launched. That means developers could start running Macron the same day it came out without waiting weeks for compatibility. That's a strong
signal. When infrastructure teams move that fast, it usually means they think There's also a smaller version called Macron tool. It's built for local deployment, meaning it can run on your own machine instead of a giant server.
That version is based on a different base model called Qwen 3.6, and it's everyday use. So, you've got two versions, a massive flagship model for heavy-duty work, and a smaller one built
because it means this isn't just for huge companies with massive computing budgets. It opens the door for smaller developers, too. Let's talk about the router itself for a second because this is the clever part. When you send a
message, a small piece called L0 looks at what you're asking and decides which specialist should handle it. If your next message is still on the same task, it stays with that specialist. But once you move to something new, it can hand
off to a different one, and it shares a short summary of what already happened, is this the model that's going to replace Claude or GPT or Gemini with you about that. On some of the harder coding and terminal benchmarks,
Mind Labs own numbers show Macron close to the top labs, but not fully past them yet. This is a strong open-source contender, not an automatic upgrade over every closed model out there. Let's bring this back to something
like this to grow a community, the way we run the AI Profit Boardroom. You could ask its agent specialist to draft a full onboarding sequence for new members. Then, hand that same context to the coding specialist to build a simple
sign-up form. Then, use the generative UI specialist to turn that into an actual working page members can click through. One system, three jobs, no re-explaining yourself each time. Macron V1 is a real signal, not a small update.
Open weights, four specialists working as one, a context window measured in millions of tokens, and infrastructure support on day one. Watch this space because if this pace keeps up, the next release might move. Here's my honest
advice. Don't try to master every new model that drops, you'll burn out. Instead, pick a system that shows you, in plain terms, how each new release actually changes what you can do in your business, and skip the rest. That's
exactly what we built the AI Profit Boardroom for, and Macron is a perfect example of why it matters. New agent tools like this are going to keep dropping every single week, and most business owners won't have the time to
test them all. Inside the Boardroom, we already do that testing for you. We run four coaching calls a week where we break down agent releases like Macron and show you exactly how to plug them into lead follow-up, content, and
customer support. You'll get daily tutorials, a 30-day roadmap built around agent tools, and a prompt library so you can start using systems like this the same day they launch instead of weeks later. Links in the comments and
description, or head to aiprofitboardroom.com. And if you want the deeper stuff, the SOPs, and over 100 real AI use cases just like this one, come join the AI Success Lab. It's free. You'll get the
notes from this exact video in there, plus access to a community of 87,000 people who are actively using AI to get ahead in their business. Links are in ahead in their business. Links are in the comments and description.