Good morning. It’s Friday, February 21st.

This Day in Tech History: In 1947, Edwin Land demonstrated the first instant camera, known as the Polaroid Land Camera.

In today’s email:

• GPT-4.5, GPT-5 Imminent Release

• Microsoft’s Quantum Chip

• Google’s PaliGemma 2

• 5 New AI Tools

• Latest AI Research Papers

GPT-4.5 and GPT-5 to be available "soon" for chat and API

OpenAI is gearing up to release GPT-4.5 and GPT-5, refining AI’s reasoning and inference game across chat and API. According to COO Brad Lightcap, GPT-5 merges OpenAI’s GPT and o-series models into a more streamlined, high-efficiency framework. Free users will get full access with no hard limits, while Plus and Pro subscribers secure extra inference time for heavier workloads.

With ChatGPT now pulling in over 400 million weekly users and two million businesses weaving it into their workflows, demand for high-caliber reasoning models has never been higher. API usage has quintupled since o3-mini’s debut, reinforcing OpenAI’s momentum. CEO Sam Altman has teased upcoming agent upgrades and "a few other things," hinting at more structural leaps ahead.

Microsoft is also preparing AI-driven Copilot enhanceme nts in tandem with the rollout, with announcements likely at its Build conference in May—coinciding with Google I/O. Read more.

Microsoft Debuts Its First Quantum Computing Chip, Majorana 1

Microsoft has introduced Majorana 1, its first quantum computing chip. The chip currently integrates eight qubits but is designed to scale to one million qubits—potentially surpassing the combined computational power of all classical computers. Microsoft’s quantum roadmap aligns with industry competitors like Google, which introduced its Willow chip in December as part of its multi-phase quantum development plan.

Despite progress, quantum computing remains in early stages. However, Microsoft is positioning Majorana 1 as a foundation for long-term breakthroughs in fault-tolerant quantum computing. Investors responded modestly, with Microsoft shares rising over 1% to $414.77. Read more.

Google’s PaliGemma 2 Mix advances vision-language AI with OCR, object detection, and captioning

Google has introduced PaliGemma 2 Mix, an expanded iteration of its PaliGemma 2 vision-language model, designed for a broader range of tasks without requiring fine-tuning. The model supports short and long captioning, optical character recognition (OCR), image question answering, object detection, and segmentation. Available in multiple sizes (3B, 10B, and 28B parameters) and resolutions (224px and 448px), it is compatible with Hugging Face Transformers, Keras, PyTorch, JAX, and Gemma.cpp.

PaliGemma 2 Mix builds on its predecessor by offering direct usability across tasks. Demonstrations include multiple object detection, OCR in various scripts, and segmentation for complex images. Users can experiment with the model via Hugging Face, Kaggle, and Google Colab, or deploy it on Vertex AI’s Model Garden. While capable out-of-the-box, fine-tuning remains an option for domain-specific applications. Documentation, sample notebooks, and inference tools are available for developers exploring the model’s capabilities. Read more.

• Biggest-ever AI biology model writes DNA on demand

• AI Agent POPPER Cuts Hypothesis Validation Time 10-Fold, Matches PhD Expertise

• Microsoft's new AI model "Muse" can generate gameplay and might preserve classic games

• Crunchbase relaunches as an AI-powered platform for predictive market insights

• Apple launches $599 iPhone 16e with Starlink satellite connectivity

• Robots learn like humans with a new framework for continuous skill acquisition

• Figure AI’s Helix arms robots with vision language and on-the-fly reasoning

• DeepSeek expands its open-source arsenal with five new AI projects

• Magma by Microsoft blends vision language and control in a single AI model

• Google’s AI ‘co-scientist’ deciphers superbug resistance in 48 hours that took scientists years

• Clone Robotics blurs the line between machine and muscle

• Nvidia helps launch AI platform for teaching American Sign Language

• Spotify partners with ElevenLabs to expand its library of AI-narrated audiobooks

• Sakana’s AI-driven CUDA tuning accelerates PyTorch up to 100x

• Microsoft’s Nadella calls AGI a distant dream while AI quietly takes over

• Microsoft opens AI research projects to developers through Azure AI Foundry Labs

• OpenAI rolls out its AI agent, Operator, in several countries

📄 Qwen2.5-VL Technical Report

📄 RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning ​​

📄 Cramming 1568 Tokens into a Single Vector and Back Again: Exploring the Limits of Embedding Space Capacity

📄 SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplines

📄 SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features

Thank you for reading today’s edition.

Your feedback is valuable. Respond to this email and tell us how you think we could add more value to this newsletter.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, reply to this email or DM us on X!