Google plans up to $40B Anthropic investment
Google is planning to invest up to $40 billion in Anthropic, combining cash and compute support. The move comes as AI companies compete to lock in huge amounts of computing power and after Anthropic’s limited release of its cybersecurity-focused Mythos model.
TechCrunch ↗
ComfyUI raises $30M at $500M valuation
ComfyUI has raised $30 million at a $500 million valuation. Its tools are aimed at giving creators more control over AI-generated image, video, and audio content.
TechCrunch ↗
Mac mini shortages fuel marked-up resale listings
Apple’s Mac mini is sold out, and resale listings on eBay are showing higher prices. The surge in demand is being driven by buyers using the compact desktop to run local AI models and tools.
TechCrunch ↗
Tim Cook to step down as Apple CEO
Tim Cook reportedly plans to leave his CEO role in September, with hardware chief John Ternus expected to take over. The transition comes as Apple faces a changing ecosystem, including pressure on the App Store’s 30% fee and shifting platform power.
TechCrunch ↗
DeepSeek previews model narrowing frontier gap
DeepSeek says its new models are more efficient and perform better than DeepSeek V3.2. The company claims they have nearly closed the gap with leading open and closed models on reasoning benchmarks.
TechCrunch ↗
Meta taps Amazon CPUs for AI workloads
Meta has reportedly signed a deal to use millions of Amazon’s homegrown CPUs for AI agentic workloads. The move suggests the AI chip race is expanding beyond GPUs as major tech firms seek alternative compute options.
TechCrunch ↗
DeepMind unveils Vision Banana image model
Google DeepMind introduced Vision Banana, an instruction-tuned image generator that the paper says outperforms leading vision models on segmentation and metric depth estimation. The article frames image-generation pretraining as a potentially important foundation for computer vision, similar to GPT-style pretraining in NLP.
MarkTechPost ↗
DeepSeek-V4 Brings One-Million-Token Contexts
DeepSeek-AI released a preview of DeepSeek-V4, a pair of Mixture-of-Experts language models designed to make one-million-token context windows more practical and affordable at inference time. The release highlights compressed sparse attention and heavily compressed attention as the key techniques behind the longer context capability.
MarkTechPost ↗
MIT opens massive Olympiad math dataset
MIT researchers have created and released a dataset of more than 30,000 Olympiad-level math problems from 47 countries. The collection is designed to give AI systems a tougher benchmark while also helping students practice advanced competition math.
MIT News ↗
Thinking Machines gains from Meta talent shifts
Meta has been hiring away talent from Thinking Machines Lab, but the movement of staff is not one-sided. The article highlights how personnel changes are affecting both companies in the AI talent market.
TechCrunch ↗