This week in AI, we’re seeing the industry redefine what’s possible across open-source, immersive technology, and funding milestones.

Switzerland launched Apertus, its first fully open multilingual AI model, trained on 15T tokens across 1,000 languages nearly half non-English. Built by EPFL, ETH Zurich, and CSCS, it sets a new standard for sovereign, transparent AI and is openly accessible via Hugging Face, Swisscom, and public utilities.

Tencent’s HunyuanWorld-Voyager generates full 3D worlds from a single image using an innovative 3D memory system. Ranked #1 on Stanford’s WorldScore, it’s open-source and designed for VR, gaming, and immersive simulations.

Anthropic raised $13B in Series F, reaching a $183B valuation. With $5B+ in annual run-rate revenue, Claude is rapidly becoming an enterprise AI leader, with fresh funding fueling global growth and safety research.

Let’s dive deeper into these breakthroughs together.

Switzerland Launches Fully Open Multilingual AI Model

Apertus is Switzerland’s first fully open, transparent, multilingual large language model, developed collaboratively by EPFL, ETH Zurich, and the Swiss National Supercomputing Centre (CSCS) as part of the Swiss AI Initiative, with support from the ETH Board and partners like Swisscom. Released on September 2, 2025, Apertus is available in two sizes (8B and 70B parameters), openly documented, and trained on 15 trillion tokens over 1,000 languages—40% non-English—to foster innovation, inclusivity, and compliance with Swiss and EU data laws. Its foundational technology, open-source license, and extensive resources enable research, education, and commercial applications, promoting responsible and sovereign AI for the public good, with practical access via Swisscom’s AI platform, Hugging Face, and the Public AI Inference Utility.

Generate Native 3D Worlds from a Single Image

HunyuanWorld-Voyager, released by Tencent’s Hunyuan AI team, is a fully open-source, ultra-long-range world model that generates native 3D reconstructions—including point clouds and RGB-D videos—directly from a single image and user-specified camera path, without traditional tools like COLMAP. It introduces an innovative 3D memory system for scalable world caching, maintaining geometric consistency across trajectories and enabling interactive control via keyboard or joystick, ranking #1 on Stanford’s WorldScore leaderboard for 3D modeling and video generation performance. Designed for VR, gaming, simulations, and spatial intelligence, Voyager empowers developers with direct 3D format exports, fully open code and model weights, and robust scene consistency for immersive, real-time AI-driven applications.

Anthropic Completes Largest Funding Round Ever

Anthropic's Claude AI has rapidly emerged as a leading intelligence platform with a $183 billion post-money valuation after raising $13 billion in a Series F funding round. Since its launch in March 2023, Claude has seen explosive growth in adoption, with run-rate revenue exceeding $5 billion by August 2025 and over 300,000 business customers worldwide. Its success is driven by cutting-edge research in AI safety, alignment, and interpretability, making it a reliable and trustworthy AI for enterprise, developer, and consumer use. Claude offers advanced capabilities such as API access, industry-specific products, and developer tools like Claude Code for coding assistance, collectively generating significant revenue. The Series F funding will expand capacity, deepen safety efforts, and support global expansion, aiming to build more interpretable and steerable AI systems for critical applications across industries.

Hand Picked Video

In this video we'll look at GPT-OSS, OpenAI's unexpected open source model that rivals O3 performance, features built-in web search, and how to test it yourself locally.

Top AI Products from this week

• Spiral – Ask our AI Agent anything across your customer reviews, support chats, calls, and surveys; get root causes by version, device, and plan - with proactive alerts before issues get expensive.

• maestro by beatoven.ai – maestro is our bold step in building an ethical music generation model. Trained on 3.5M licensed tracks, this foundation model is capable of crafting high quality music with text prompts. Your personal AI composer is now more creative and inspirational.

• Incerto – Incerto is an AI-native Copilot for databases. Using context-aware agents, it turns natural language into tasks — like writing queries, schema changes, or moving data — boosting dev productivity 10x with 90% less manual work.

• Mascotly - Mascotly helps you turn your brand into a character people actually remember. No more soulless logos - generate fun, cute, and unique mascots with AI in seconds. Perfect for startups, creators, and businesses that want to stand out.

• Shotgun CLI – Your AI posse for researching, planning & building software. Repo-aware, multi-agent, and local-first — Shotgun creates decision-ready specs and shareable artifacts that truly capture your intents and context, so AI tools finally build what you mean.

• AI OmniBrief – A weekly meta-newsletter that cuts through AI noise. Every Monday, it delivers the most important research, models, tools, deployments, industry moves, and governance updates. Curated, fact-checked, and linked to primary sources.

• Simulai - Test your AI agents with AI users across different personas and scenarios. Catch issues before they hit your real users.

This week in AI

• Amazon Lens Live - AI-Powered visual search instantly scans products, shows real-time matches in a swipeable carousel, and integrates AI assistant Rufus for insights. Available now on iOS in the US.

• ElevenLabs SFX v2 Update - ElevenLabs boosts SFX to 30s, 48kHz, bigger library, Remix + Favorites, upgraded SB-1 soundboard with seamless loops & MIDI—perfect for film, TV, games.

• LongCat-Flash-Chat - Meituan’s 562B-param MoE LLM activates ~27B params per query, enabling 100+ TPS, 128k context, strong reasoning, coding & agentic tasks; open-sourced under MIT.

• R-4B Auto-Thinking Model - OpenAI is adding ChatGPT parental controls to link parent-teen accounts, limit features, and alert parents during teen distress. Sensitive chats route to safer reasoning models.

• Mistral AI Memories Feature - Mistral AI's Le Chat stores and recalls past interactions with transparency and user control. It personalizes chats, supports edits, citations, and data sovereignty, enhancing workflow continuity.

Paper of The Day

SAM-LLM: Interpretable Lane Change Trajectory Prediction via Parametric Finetuning" proposes a hybrid model combining large language model (LLM) reasoning with kinematic lane change modeling. Instead of predicting raw trajectory coordinates, the LLM predicts physically meaningful parameters (lateral displacement, maneuver duration, initial lateral velocity, longitudinal velocity change) of an enhanced Sinusoidal Acceleration Model (SAM). This approach reduces output size by 80%, improves computational efficiency with 54% faster inference, and provides interpretable, smooth, physically plausible trajectories. The SAM-LLM achieves state-of-the-art lane change intention accuracy (98.73%) on the highD dataset, offering robust and efficient predictions while bridging AI reasoning with autonomous driving.

To read the whole paper, go to here.