Three reasons why DeepSeek's new model V4 matters

The long-awaited V4 model is more efficient and a win for Chinese chipmakers.

Meta's loss is Thinking Machines' gain

Meta has been poaching talent from Thinking Machines Lab. But it's a two-way street.

SoftBank is converting a Sharp LCD factory into a battery plant for AI data centres. The data centres cannot wait five years.

Summary: SoftBank Corp. plans to convert part of the former Sharp LCD factory in Sakai, Osaka into one of Japan’s largest battery production lines for AI data centres, with production expected within five years (~2031). The move completes a vertical integration stack spanning chips (Arm, Graphcore, Ampere), modular data centre manufacturing (Lordstown), energy generation (SB Energy, 3+ GW solar), and now energy storage. SoftBank carries ~$135 billion in total debt with an S&P negative outlook, and its Stargate AI infrastructure commitments will require the batteries long before the factory is operational.

ComfyUI Hits $500M Valuation as Creators Seek More Control Over AI-Generated Media

ComfyUI, whose tools give creators more control over AI image, video, and audio generation, just raised $30 million.

Google to invest up to $40B in Anthropic in cash and compute

Google plans up to $40B investment in Anthropic as AI rivals race to secure massive compute capacity, following the limited release of its powerful, cybersecurity-focused Mythos model.

Apple's new CEO, and why Elon Musk wants to buy Cursor for $60B

Equity digs into what Apple's leadership transition means for startups and more of the week’s biggest deals, including SpaceX's $60B option on Cursor.

Report: Samsung execs worried company could lose money on smartphones for the first time

The AI-driven memory shortage is hitting Samsung’s bottom line.

Marked-up Mac minis flood eBay amid shortages driven by AI

Apple’s sold-out Mac mini is spawning marked-up eBay listings as demand surges for the compact desktop, now favored for running local AI models and tools.

Submodular Optimization for Text Selection, Passage Reranking & Context Engineering

After my previous article on submodular optimization for fan-out queries in DeepResearch, I received great feedback asking for a deeper dive into submodularity especially in information retrieval and agentic search. Today, I'll introduce two more applications of submodular optimization: text selection and passage reranking. Both tackle the same core challenge—optimal subset selection—that every DeepResearch-like system must solve.

Submodular Optimization for Diverse Query Generation in DeepResearch

Learn more from the part 2 of the submodularity series. More applications of submodular optimization on text selection, passage reranking and context engineering.

jina-reranker-m0: Multilingual Multimodal Document Reranker

Today we're releasing jina-reranker-m0, our new multilingual multimodal reranker model for ranking visual documents across multiple languages: it accepts a query alongside a collection of visually rich document images, including pages with text, figures, tables, infographics, and various layouts across multiple domains and over 29 languages. It outputs a ranked list of documents ordered by their relevance to the query. Compared to jina-reranker-v2-base-multilingual, jina-reranker-m0 also improves text reranking for multilingual content, long documents, and code searching tasks.

JinaVDR: New Visual Document Retrieval Benchmark with 95 Tasks in 20 Languages

We're releasing JinaVDR (Visual Document Retrieval), a new benchmark for evaluating how well models retrieve visually complex documents. JinaVDR encompasses multilingual documents with intricate layouts—combining graphs, charts, tables, text, and images alongside scanned copies and screenshots. The benchmark pairs these diverse visual documents with targeted text queries, enabling comprehensive evaluation of retrieval performance across real-world document complexity and broader domain coverage.