Data Points: Swiss model Apertus discloses code, datasets

Welcome back! In today’s edition of Data Points, you’ll learn more about:

Hermes, a model trained to reason and follow instructions
Tencent’s new explorable 3D world framework
OpenAI’s new jobs and certification programs
Warner Bros’s lawsuit against Midjourney

But first:

Swiss universities unveil fully open source AI model

Switzerland launched Apertus, a national LLM developed by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre as an alternative to ChatGPT, Meta’s Llama models, and DeepSeek. The model comes in two sizes — 8 billion and 70 billion parameters — and was trained on 15 trillion tokens across more than 1,000 languages. 40 percent of Apertus’s training data is non-English, including underrepresented languages like Swiss German and Romansh. Unlike commercial models, Apertus’s (from the Latin for “open”) architecture, model weights, training data, and development recipes are all openly accessible and fully documented, ensuring compliance with Swiss data protection laws and EU AI Act transparency requirements. The models are freely available under a permissive open source license for educational, research, and commercial applications, with deployment supported through platforms like Transformers, vLLM, SGLang and MLX. (Swiss AI)

China enforces new AI labeling requirements across major platforms

China’s new law requiring labels for all AI-generated content took effect this week, prompting social media platforms including WeChat, Douyin, Weibo, and Xiaohongshu to implement compliance features. The regulation mandates both explicit visible labels and implicit metadata identifiers for AI-generated text, images, audio, video, and other synthetic content. Platforms now offer tools for creators to declare AI-generated content voluntarily, while also deploying detection systems to identify unlabeled AI material. The law reflects Beijing’s growing concerns about misinformation, copyright infringement, and online fraud related to AI technology, particularly deepfakes. The regulation was drafted by China’s Cyberspace Administration along with three other government ministries as part of the country’s broader AI oversight efforts. (South China Morning Post)

Nous Research releases Hermes 4 open-weight reasoning models

Hermes 4 models combine structured reasoning capabilities with broad instruction-following abilities, reducing reasoning chains and offering more versatility. In training the model, the team developed specialized techniques including DataForge for synthetic data generation and length-control fine-tuning to manage excessive reasoning lengths. Hermes 4 405B achieved 70.6 percent on GPQA-Diamond and 81.9 percent on AIME 24 in reasoning mode (outperforming Qwen 3 and DeepSeek V3, falling just short of DeepSeek-R1) while maintaining strong performance on general benchmarks. All model weights are publicly available at Hugging Face. (arXiv)

Tencent’s Voyager creates explorable 3D worlds from images

Tencent released Voyager, a new AI system that creates 3D environments from just one image, letting users navigate through generated scenes along custom camera paths. The framework generates both color video and depth information simultaneously, allowing it to build 3D scenes directly without needing complex reconstruction steps required by other methods. The system uses three main parts: a video generator that keeps scenes consistent as the camera moves, a memory system that stores previously generated areas for smooth exploration, and a data processing tool that prepares training videos automatically. This system could make it easier to create virtual worlds for games, movies, and robot training, since it removes the need for manual 3D modeling work that traditionally takes significant time and expertise. Voyager is available on HuggingFace and GitHub. (Tencent)

OpenAI launches job platform and certification program

OpenAI launched two new workforce initiatives aimed at training 10 million Americans in AI skills by 2030, partnering with companies including Walmart, John Deere, and Accenture. The company’s jobs platform will connect workers with AI skills to employers seeking such talent, while certifications will be offered through ChatGPT’s Study mode for various skill levels from basic AI use to prompt engineering. The company states these initiatives are part of its commitment to the White House’s AI literacy efforts, though specific details about program costs, certification standards, and job placement outcomes remain unclear. (OpenAI)

Warner Bros. sues Midjourney for copyright infringement

Warner Bros. filed a federal lawsuit against image generation company Midjourney, claiming the startup allows users to generate unauthorized images and videos of copyrighted characters including Superman, Batman, and Bugs Bunny. The lawsuit alleges Midjourney trained its AI system on “illegal copies” of Warner Bros. works and that even generic prompts like “classic comic book superhero battle” produce images of DC characters. Warner Bros. seeks up to $150,000 in damages per infringed work and argues Midjourney could implement content restrictions similar to its existing limits on violence and nudity. This marks the third major Hollywood studio to sue Midjourney, following a joint lawsuit by Disney and Universal in June. (Associated Press)

Still want to know more about what matters in AI right now?

Read this week’s issue of The Batch for in-depth analysis of news and research.

This week, Andrew Ng wrote about the growing unmet demand for AI-skilled developers, the challenges recent computer science graduates face in the job market, and why combining strong fundamentals with modern AI tools is key to thriving as a developer today.

“There is significant unmet demand for developers who understand AI, while recent CS graduates face increased unemployment because most universities have not yet adapted their curricula to the new reality.”

Read Andrew’s full letter here.

Other top AI news and research stories we covered in depth:

Chatbot interviewers are helping companies fill more customer service roles, with studies showing improvements in both hiring and retention.
In China, DeepSeek and other “little dragons” are turning Hangzhou into a rising AI hub often called the country’s Silicon Valley.
Google published a direct measurement of Gemini’s environmental impact, detailing electricity, water use, and greenhouse emissions.
Meta introduced LlamaFirewall, an open source tool designed to protect AI agents against hijacking attacks.

Subscribe to Data Points