🚀 OpenAI releases o3-pro 🚀
Performance available to ChatGPT Pro users and API

OpenAI has released o3-pro with significantly better performance than o3, now available to ChatGPT Pro users and through the API. The model achieves new records on the Extended NYT Connections benchmark (87.3 up from 82.5) and has become the #1 model on SnakeBench. In a remarkable demonstration of its capabilities, o3-pro one-shotted the Tower of Hanoi 10 disk problem. Additionally, OpenAI has doubled the rate limits for o3 for ChatGPT Plus users and introduced fine-tuning capabilities for the GPT-4.1 family using direct preference optimization (DPO).

🤖 Databricks Agent Bricks 🤖
New approach to building auto-optimized agents

Databricks has introduced Agent Bricks, a declarative and compositional approach to building AI agents. The platform allows developers to use natural language feedback to steer agents toward optimal performance, simplifying the creation of complex agent systems while maintaining control over their behavior and optimization.

🧠 Mistral AI announces Magistral 🧠
First specialized reasoning model with multilingual capabilities

Mistral AI has officially announced Magistral, their first reasoning model designed for domain-specific, transparent, and multilingual reasoning. The 24B parameter model is based on Mistral Small 3.1 and offers a 128K context length under the Apache 2.0 license. For developers looking to implement it immediately, Magistral 4-bit DWQ is available on Hugging Face for use with mlx-lm or LM Studio. This release represents a significant step forward in Mistral's specialized model offerings, particularly for applications requiring advanced reasoning capabilities across multiple languages.

INTERESTING NEWS

⚖️ Disney and Universal sue Midjourney ⚖️

Disney and Universal have filed major lawsuits against AI image generator Midjourney, accusing it of widespread copyright infringement and describing the platform as a "bottomless pit of plagiarism." The case centers on the use of copyrighted images in training data without explicit licensing. This legal battle could set key precedents affecting all AI labs using similar scraped data, with the core debate focusing on whether training generative models on copyrighted works constitutes fair use.

☁️ OpenAI partners with Google Cloud for additional compute ☁️

OpenAI has tapped Google for a new cloud deal to secure additional computing resources, diversifying beyond its primary partnership with Microsoft. This strategic move comes as AI companies race to secure the necessary infrastructure for training and running increasingly powerful models. In related infrastructure news, Mistral AI unveiled Mistral Compute as an "unprecedented AI infrastructure undertaking in Europe," while NVIDIA and Hugging Face announced a collaboration to connect AI researchers with GPU clusters via a new Training Cluster initiative. These developments highlight the critical importance of computing resources in the AI arms race.

H-FARM EVENTS ON AI

H-FARM CLUB Artificial Intelligence & Robotics
Ready to explore how technology is reshaping our world?

Join our AI Club and be part of the conversation in this special edition as we dive into AI, humanoid robots, creative innovation, and their concrete implications for the future.
Artificial intelligence keeps evolving at a rapid pace, but it’s robotics, now more intelligent and accessible than ever, that makes this transformation truly tangible.
June 17 at 6:00 PM at H-FARM Campus, we'll take you to the heart of this revolution, from software to matter, from code to movement.

Our guests:

• Diego Pizzocaro, CEO of H-FARM AI, with the latest AI news and insights

• Jakidale (Content Creator) & Marco Savini (Founder of BigRock) presenting "Robot in Action: AI, Creativity & Movement"