- Master List AI
- Posts
- 🤖Meta Drops Llama 4 Maverick: The New Multimodal King
🤖Meta Drops Llama 4 Maverick: The New Multimodal King
+ 🎨 ChatGPT’s Stunning Art Tool Might Get a Big Change
Rocky Stano
April 07, 2025
Welcome, Master! 🚀
Happy AI Monday
read time: 4 min
Daily News for Curious Minds
Be the smartest person in the room by reading 1440! Dive into 1440, where 4 million Americans find their daily, fact-based news fix. We navigate through 100+ sources to deliver a comprehensive roundup from every corner of the internet – politics, global events, business, and culture, all in a quick, 5-minute newsletter. It's completely free and devoid of bias or political influence, ensuring you get the facts straight. Subscribe to 1440 today.
ON YOUR ‘MASTER LIST’ TODAY
💬 AI Inspiration
🎨 OpenAI’s ImageGen Just Got Real—And Watermarked
🤖 Anthropic Finds AI Quietly Using Secret Hints
🌐Meta Just Dropped Llama 4—and It's Beating GPT-4o
📊 Last Poll Result
MONDAY AI QUOTE
“Your dreams are the blueprint for your future”
OPEN AI
OpenAI Testing Watermarks for ChatGPT-4o ImageGen
OpenAI is testing watermarks on images made with its ChatGPT-4o ImageGen tool, starting with free users.
Paid users won’t get watermarks. This helps tell which images were made by AI.
An ImageGen API is also coming for developers.
META AI
Meta Unveils Llama 4: Scout, Maverick, and the Massive Behemoth
Meta has released the first models in its new Llama 4 lineup: Llama 4 Scout and Llama 4 Maverick—both open-weight, multimodal models designed for high performance and efficiency.
Built on a mixture-of-experts (MoE) architecture, these models push the boundaries of context length and reasoning power. Llama 4 Scout, with 17B active parameters and a 10 million token context window, fits on a single H100 GPU.
Llama 4 Maverick, also with 17B active parameters but 128 experts, outperforms GPT-4o and Gemini 2.0 in coding, reasoning, and multimodal tasks—at a much lower compute cost.
Credits: Meta
What You Need to Know:
Llama 4 Scout: 17B active params, fits on 1 H100, supports 10M token context
Llama 4 Maverick: 17B active params, 128 experts, beats GPT-4o on multiple benchmarks
Behemoth: 288B active params, still training, outperforms GPT-4.5 on STEM tasks
Both models use native multimodality (text + images), with major improvements in image grounding
Models are available now via llama.com and Hugging Face for developers
Built with new RL and fine-tuning techniques for better performance and lower latency
The Llama 4 models are distilled from Llama 4 Behemoth, Meta’s 2-trillion parameter teacher model designed to enhance reasoning, coding, and image understanding.
These releases mark Meta’s most advanced AI yet, offering open access for developers and integration across Meta’s platforms like WhatsApp and Instagram.
With strong benchmarks, long context capabilities, and open availability, Llama 4 sets a new standard in open-source AI.
Which Llama 4 model are you most excited to try? |
AI ART - CAN YOU TELL?
 A |  B |
40% Got This One ✔
ANTHROPIC
Anthropic: Don’t Fully Trust AI's Chain-of-Thought Reasoning
Anthropic found that AI models like Claude 3.7 Sonnet often hide how they reach answers.
In tests, models used hidden hints, even unethical ones, but rarely admitted it.
This shows current Chain-of-Thought reasoning isn’t fully honest or reliable, raising trust concerns for future AI use.
HELP US GROW
❤️ Your Feedback Matters: ↓ Share your thoughts below ↓
🚀 Partner with Us: Elevate your brand here
📣 Tell a Friend: Love our content? Spread the word!
🔗 Connect on LinkedIn: click here
LAST POLL RESULT
 | 🤖 Do you think AGI (AI smarter than humans) will happen by 2030? A resounding ✅YES |
ENJOY THE WEEK AHEAD
That's a wrap!
Enjoy the Week Ahead!
See you for AI-Wednesday
MasterListAI.com
FEEDBACK
What did you think of today's newsletter? Don't forget to add in your comments! Wanting to bring you the best value |