🚀 Claude plans first AI-guided drive on Mars 🚀
NASA's Perseverance executes 400m route plotted by Anthropic's AI

NASA revealed that its Perseverance rover completed the first AI-planned drive on another planet, with Claude Code mapping a 400-meter route across the Martian surface that the rover navigated in December. Engineers fed Claude years of rover driving data, with the AI writing navigation commands and plotting waypoints around rocks and sand ripples. The team verified routes through simulation before transmission to Mars, requiring only minor changes. AI-assisted planning could cut route-mapping time in half, enabling additional drives and faster data collection.

🎬 Grok Imagine tops video generation leaderboards 🎬
Musk's AI video model debuts at #1 with aggressive pricing

xAI released the Grok Imagine API for video generation, immediately climbing to #1 on Artificial Analysis rankings for both text-to-video and image-to-video outputs. The model generates 15-second clips with native audio at $4.20 per minute, significantly undercutting competitors like Veo 3.1 ($12/min) and Sora 2 Pro ($30/min). The API includes advanced editing tools for object swapping, scene restyling, and custom character animations. Elon Musk claims Grok Imagine now generates more images and videos than all competitors combined.

🤝 Google tests Claude in Gemini for Business 🤝
Third-party models coming to Google's AI platform

Google is testing the integration of Anthropic's Claude Sonnet 4.5 into its Gemini for Business platform, marking a significant shift in strategy. The move would give enterprise users access to multiple AI models from a single interface, potentially streamlining workflows and reducing vendor lock-in. This follows the broader industry trend of model interoperability, with similar integrations appearing across major AI platforms.

INTERESTING TO KNOW

🤖 NVIDIA launches Cosmos Policy for robot control 🤖

NVIDIA released Cosmos Policy, a robot control system that post-trains the Cosmos Predict-2 world foundation model for manipulation tasks. The model treats robot actions, physical states, and success scores just like video frames, enabling it to predict action chunks, future observations, and expected returns simultaneously. Cosmos Policy adapts through a single training stage on robot demonstration data, eliminating the need for separate models for different robot tasks.

🇨🇳 How Chinese AI labs build open models 🇨🇳

Chinese labs move fast through first-principles thinking, engineering discipline, and willingness to work whenever the model requires them to, according to MiniMax senior researcher Olive Song. The interview reveals how cutting-edge AI research is actually done inside a Chinese lab, covering topics like alignment, training challenges, agentic RL, and the relationship between coding and general intelligence. The practical, hands-on approach enables rapid iteration.