Tenant Releases An Open-Source Native Multimodal Model for Image Generation

Hey there 👋

We hope you're excited to discover what's new and trending in AI, ML, and data science this week.

Here is your 5-minute pulse...

The best way to future proof your Data Science career

This bootcamp helps data professionals master AI workflows and automation so they can stay relevant, work smarter, and accelerate career growth.

Over 5 weeks, you’ll:

• Go beyond just prompting ChatGPT

• Build real AI workflows that save hours

• End with your own Slackbot that talks to your data

📌 Enrollment for the October 2025 closes in a week

Secure your spot (only a few more left) 

print("News & Trends")

Tencent Releases HunyuanImage-3.0: An Open-Source Native Multimodal Model for Image Generation (7 min. read)

Tencent's HunyuanImage-3.0 is a cutting-edge multimodal model designed for high-quality image generation. It leverages advanced diffusion transformers and supports both Chinese and English prompts, enabling seamless text-to-image synthesis. The model's architecture integrates a multimodal large language model to enhance image-text alignment and a character-aware encoder to improve text rendering across various languages. With its efficient two-stage process—a base text-to-image model followed by a refiner model—HunyuanImage-3.0 produces ultra-high-definition (2K) images with remarkable clarity and detail. This open-source release offers data scientists and ML engineers a robust tool for exploring and advancing image generation technologies.

OpenAI Updates API Function Calling To Support Files and Images (5 min. read)

OpenAI's Function Calling Guide introduces a powerful feature that enables models to generate structured outputs, facilitating seamless integration with external APIs and databases. This functionality allows developers to define functions that the model can call, enhancing its ability to perform specific tasks like retrieving real-time data or executing operations beyond text generation. By leveraging function calling, you can create more dynamic and interactive applications, bridging the gap between AI models and practical, real-world use cases.

Former Microsoft Execs Launch AI Agents to End Excel-Led Finance (4 min. read)

Two ex-Microsoft leaders have unveiled Maximor, an AI-driven platform designed to replace Excel in financial operations. By integrating directly with ERP, CRM, and billing systems, Maximor's AI agents continuously pull and reconcile data, offering real-time financial insights. Early adopter Rently reduced its month-end close from eight days to four and reallocated nearly half its team's time to strategic tasks. While aiming to diminish Excel dependence, Maximor still allows data exports to spreadsheets, catering to traditional preferences.

Apple's Veritas: A ChatGPT-Like Leap for Siri (3 min. read)

Apple is reportedly developing 'Veritas,' an AI-powered voice assistant poised to revolutionize Siri by integrating ChatGPT-like capabilities. This move aims to enhance user interactions, making them more natural and context-aware. While details remain under wraps, Veritas could signify a significant leap in Apple's AI strategy, positioning Siri as a more formidable competitor in the evolving landscape of intelligent voice assistants.

print("Applications & Insights")

Jupyter Agents: Training LLMs to Reason with Notebooks (5 min. read)
Hugging Face introduces Jupyter Agent 2, a model designed to execute code within Jupyter notebooks, enhancing data analysis tasks. By fine-tuning smaller models with a novel dataset derived from real Kaggle notebooks, they achieved a significant performance boost on the DABStep benchmark, with accuracy on easy tasks rising from 44.4% to 70.8%. This advancement underscores the potential of integrating code execution capabilities into language models for more effective data science applications.

I Made My AI Model 84% Smaller and It Got Better, Not Worse (20 min. read)
This article shows how dynamic INT8 quantization cut model size by 84% without hurting accuracy, enabling fast edge deployment with lower costs and latency. By combining domain-adaptive pretraining, compression, and a smart router that sends only complex queries to the cloud, the author built a hybrid AI system that delivers cloud-level performance at edge-level efficiency .

AI-Generated “Workslop” Is Destroying Productivity (5 min. read)
Despite the surge in AI adoption, a staggering 95% of organizations report no measurable ROI from these technologies. The culprit? An influx of low-quality, AI-generated content—dubbed "workslop"—that clogs workflows and hampers productivity. This article delves into the paradox of AI enthusiasm versus its underwhelming impact, urging a strategic reassessment to harness AI's true potential.

The Evolution of AI: The State of Enterprise AI and Data Architecture (7 min. read)
Cloudera's latest survey reveals that 96% of enterprises have integrated AI into core business processes, marking a shift from experimentation to essential practice. A significant 70% report substantial success with AI initiatives. Organizations are leveraging various AI forms, including generative (60%), deep learning (53%), and predictive models (50%). The adoption of hybrid data architectures is prevalent, with 63% utilizing private clouds and 52% public clouds, emphasizing the need for flexible, secure, and scalable AI solutions.

Video Models Are Zero-Shot Learners and Reasoners (3 min. read)
Veo 3, a generative video model, exhibits impressive zero-shot capabilities across diverse visual tasks—ranging from object segmentation and edge detection to understanding physical properties and simulating tool use. These emergent abilities suggest that video models are evolving into generalist vision foundation models, paralleling the trajectory of large language models in natural language processing.

print("Tools & Resources")

TRENDING MODELS

Text-to-Image
tencent/HunyuanImage-3.0
⇧ 330 Downloads
HunyuanImage-3.0 is a text-to-image model developed by Tencent, capable of generating high-quality images from textual descriptions. It features 83 billion parameters.

Any-to-Any
Qwen/Qwen3-Omni-30B-A3B-Instruct
⇧ 125k Downloads
Qwen3-Omni-30B-A3B-Instruct is a versatile model designed for various tasks, including text generation and understanding.

Text Generation
deepseek-ai/DeepSeek-V3.2-Exp
DeepSeek-V3.2-Exp is a text generation model by DeepSeek AI, featuring 685 billion parameters.

Image-to-Image
Qwen/Qwen-Image-Edit-2509
⇧ 25.1k Downloads
Qwen-Image-Edit-2509 is an image-to-image model developed by Qwen, designed for advanced image editing tasks.

Text-to-Speech
openbmb/VoxCPM-0.5B
⇧ 5.47k Downloads
VoxCPM-0.5B is a text-to-speech model by OpenBMB, capable of generating human-like speech from text input.

TRENDING AI TOOLS

• 👥 HumanLayer: AI-powered tool for understanding and managing human-centric data efficiently.

• 🔍 RAGLight: Lightweight framework for retrieval-augmented generation in NLP tasks.

• 🚀 Kilo Code for JetBrains: AI-powered code generation and completion for JetBrains IDEs.

• 🎓 Creatium: Build interactive video lessons and adaptive AI video coaches to enhance online learning experiences.

print("Everything else")

• GitHub now offers the Copilot coding agent generally to paid Copilot users, letting it autonomously open draft pull requests and work in the background.

• ChatGPT provides parental controls so parents can link teen accounts, set limits, and adjust content safeguards.

• OpenAI has dropped their “ChatGPT usage and adoption patterns at work” report.

• Meta poaches yet another OpenAI Scientist to lead part of its AI research lab.

That’s it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve the pulse experience for you.

See you soon,

Andres