- Horizon AI
- Posts
- Llama 4 is Here 🦙
Welcome to another edition of Horizon AI,
Meta just dropped a powerful new family of models over the weekend, featuring a massive context window and new multimodal capabilities. Meanwhile, OpenAI announced a change of plans for the release of its next models.
Let’s jump into it!
Read Time: 4.5 min
Here's what's new today in the Horizon AI
Meta Releases Two Llama 4 AI Model
OpenAI Shifts Course, GPT-5 Delayed
AI Tutorial: Clone any writing style using AI
AI Tools to check out
The Latest in AI and Tech 💡
AI Findings/Resources
AI News
META
Meta Releases Two Llama 4 AI Model
Meta has released the first two models in its new Llama 4 series that the company claims are superior to those from OpenAI and Google across ‘a broad range’ of benchmarks.
Details:
The two models use a Mixture-of-Experts (MoE) design, in which only a subset of parameters is activated for each input, reducing computational overhead.
The 109B parameter Llama 4 Scout features a 10-million token context window. It is optimized to run on a single H100 GPU and is designed for tasks such as long-form text processing, visual question answering, code analysis, and multi-image understanding.
The 400B parameter Llama 4 Maverick supports a context window of up to one million tokens, offering superior performance while being more efficient.
According to Meta, Llama 4 Scout beats Gemini 3, Gemini 2.0 Flash-Lite models, and Mistral 3.1 “across a broad range of widely reported benchmarks,” while Llama 4 Maverick outperforms GPT-4o and Gemini 2.0 Flash.
Meta says it’s still in the process of training Llama 4 Behemoth, a nearly two-trillion-parameter model, which would be “the highest performing base model in the world,” outperforming its competitors (GPT-4.5 and Claude Sonnet 3.7) “on several STEM benchmarks.”
Both Scout and Maverick are available as open-weight models via llama.com and Hugging Face, and they now power the Meta AI assistant on the web and in WhatsApp, Messenger, and Instagram. You can try them in each of those.
TOGETHER WITH HEAR.COM
Forget everything you know about hearing aids
Featuring one of the world's first dual-processing systems that separates speech from background noise, ensuring you never miss a beat, a word, or a moment. Claim your free consultation and join the 400k+ who can hear with crystal clarity thanks to these game-changing hearing aids.
OPENAI
OpenAI Shifts Course, GPT-5 Delayed
Sam Altman announced a revised roadmap for new AI models, as the company found that integrating all components into a single unified system (GPT-5) proved more challenging than initially anticipated.
Details:
OpenAI now plans to release both o3 and a next-gen successor, o4-mini, within “a couple of weeks.”
The full GPT-5 model is now expected to arrive "in a few months."
According to Altman, the change is due to multiple factors, including integration challenges and the need to ensure enough computing capacity to meet what is expected to be "unprecedented demand."
OpenAI had originally planned for GPT-4.5 to be the final model in its line of conventional language models. GPT-5 was intended to merge the GPT and o-model series into a single platform, combining tools such as natural language processing, search, and deep research into one system. Altman described it as a "magic unified intelligence" capable of automatically selecting the appropriate tool for a given task.
AI Tutorial
Clone any writing style using AI
Go to Claude’s website and sign up.
Click ‘Choose Style,’ then select ‘Create and Edit Styles.’
Click ‘Create Custom Style.’
Upload a writing sample in your preferred style so Claude can analyze it properly.
Click ‘Create Style’ and wait for it to process.
Once it’s ready, click ‘Use Style’ to start using it.
Claude will respond in the writing style you provided within seconds.
AI Tools to check out
⏰ Clockwise: AI calendar assistant that helps busy people schedule meetings and automate their calendars.
🚀 Pitchleague: Elevate your pitch with your AI pitch deck coach.
✨ CreateMyBanner: Transform your ideas into eye-catching banners instantly.
📝 Tana: Put your notes to work with voice and AI
💡 Marketing Ideas Generator: Get 10 actionable ideas to solve your marketing problem.
AI Findings/Resources
📚 Microsoft is offering free AI skills training for everyone - how to sign up
📷 Can ChatGPT 4o image generation REALLY generate product photo ads?
👉 Run AI models locally with Ollama: Fast & simple deployment
😱 AI isn’t what we should be worried about – it’s the humans controlling it
The latest in AI and Tech
Scott Alexander and former OpenAI researcher Daniel Kokotajlo released a deeply researched, month-by-month scenario predicting the monumental event of artificial intelligence overtaking human intelligence by 2027, urging immediate action on AI safety and ethical guidelines.
It is built on Microsoft Muse, a generative AI model it released earlier in 2025, which can replicate existing games based on its training data.
The tech demo is playable in a browser and free to try, but Microsoft also appears to have limited how long you can play.
The celebration was interrupted by multiple protesters on Friday due the military use of the company's AI technology.
The study, co-authored by researchers from the University of Washington, the University of Copenhagen, and Stanford, proposes a new method for identifying training data "memorized" by models behind an API, such as OpenAI’s.
It found that the company’s models showed signs of having memorized portions of popular fiction books and New York Times articles.
While AI is transforming economies and creating new opportunities, it also risks deepening existing inequalities, the UN warned in a report.
In particular, the report cautioned that "AI could impact 40 percent of jobs worldwide, offering productivity gains but also raising concerns about automation and job displacement".
That’s a wrap!
We'd love to hear your thoughts on today's email!Your feedback helps us improve our content |
Not subscribed yet? Sign up here and send it to a colleague or friend!
See you in our next edition!
Gina 👩🏻💻