Technology

Xiaohongshu and Peking University have released UltraEP research that challenges a basic assumption behind large-scale mixture-of-experts infrastructure. Instead of predicting which experts will become busy, UltraEP MoE load balancing reacts after each layer’s router reveals the actual demand. That distinction matters because expert demand can shift between microbatches, layers, and prompt domains. A placement plan based on recent history can become stale before the system us

The Shanghai Academy of AI for Science has opened ShenZhen, an approximately 11-billion-parameter model built to handle six scientific data types. The Shanghai Academy of AI for Science ShenZhen multimodal foundation model covers DNA, RNA, proteins, small molecules, Earth systems, and medical images. That scope creates the central tension. Scientific AI has usually advanced through specialized models designed around one domain, data structure, or prediction target. ShenZhen,

Alibaba's Qwen team released Qwen-Audio-3.0-TTS with two versions, one promising near-instant output and another already leading a major voice benchmark. The Qwen-Audio-3.0-TTS real-time speech synthesis model supports 16 languages and 20 Chinese dialects, according to the company. That combination matters more than either headline alone. Voice developers usually choose between fast responses, expressive delivery, accurate pronunciation, and consistent speaker identity. Aliba

Ollama has secured $65 million in Series B financing, bringing its total funding to $88 million as it expands beyond local AI development. The Ollama $88 million funding story is not simply another large check for an AI infrastructure company. It puts new pressure on Ollama to connect local computing with cloud capacity without weakening its ownership and privacy message. Theory Ventures led the new round, with participation from Benchmark, 8VC, Y Combinator, Pace Capital, 49

Nvidia CEO Jensen Huang used a two-day Japan visit to announce a 140-megawatt AI factory and recruit the country’s industrial leaders into a shared robotics effort. The Nvidia Japan physical AI push is not simply another data center order. It connects national computing capacity, domestic industrial data, open model development, and factory deployment through one Nvidia-centered technology stack. Noetra plans to install approximately 27,500 Rubin GPUs and 13,750 Vera CPUs, wi

Kimi has turned a once technical workflow into a conversational one, but the shift creates a new conflict between building quickly and shipping responsibly. This Kimi vibe coding tutorial examines that conflict through a complete product journey, from an initial specification to a public deployment. The important change is not that an AI model can generate a landing page. Coding agents can now inspect project files, edit several components, execute commands, run tests, and ad

ModelBest and OpenBMB released MiniCPM-Robot on July 19, their first open model family built to perceive, decide, and act in the physical world. The release combines a 1.5-billion-parameter manipulation model with a 0.9-billion-parameter target tracker. That pairing creates a more interesting contest than another benchmark race. MiniCPM-Robot tests whether small, open models can deliver useful robotic behavior without depending on large servers, proprietary access programs, o

ByteDance has released Seed Audio 1.0 with a direct challenge to conventional AI audio production. One model now generates voices, music, ambience, and effects together. The distinction matters because most generative audio workflows remain divided across specialized systems. A creator generates dialogue, adds music elsewhere, sources effects, aligns every element, and mixes the final timeline. ByteDance Seed Audio 1.0 attempts to collapse that chain into a single prompt-driv

Qwen announced a 2.4-trillion-parameter model and promised open weights, but the downloadable release has not arrived. The Qwen 3.8 open-weight model announcement instead directs users to a hosted preview inside several Alibaba products. Qwen says Qwen3.8-Max-Preview is available through Token Plan, Qoder, and QoderWork. Users can test an Alibaba-controlled service, yet they cannot inspect or independently deploy the promised model. The claimed scale makes the announcement di

Red Hat eliminated its Beijing and Shanghai engineering teams in mid-2026, transferring work to India centers while citing regulatory overhead and cost efficiency.

MiniCPM5-2B has arrived with a striking claim: the 2-billion-parameter model leads every model below 4 billion parameters on the AA Index. The MiniCPM5-2B release also promises native 512K context, selectable thinking modes, and Day 0 adaptation across nine chip platforms. ModelBest and OpenBMB presented those features as one package built for local and edge deployment. That combination matters more than another small-model benchmark win. Qwen3.5-2B already offers hybrid reas

Ludicity published a blunt warning on July 18, 2026: AI mania is eviscerating global decision-making, despite mounting evidence that many enterprise projects never deliver measurable value. The essay is not another prediction about an approaching AI bubble. It describes institutions already changing how they approve projects, hire employees, evaluate performance, and report failures. Its central allegation is that leaders increasingly reward visible AI adoption over useful ou

OpenBMB has announced its first embodied AI series, built around a 1.5-billion-parameter manipulation model and a separate tracking model. The OpenBMB MiniCPM-Robot embodied AI model series targets a difficult conflict: making robot intelligence smaller and more accessible without sacrificing dependable physical performance. The company’s MiniCPM-Robot announcement names three components. MiniCPM-RobotManip handles vision, language, and robot actions. MiniCPM-RobotTrack follo

Kunlun Tech declared 2026 the “year of world models” and introduced Matrix-Game 3.5 at the World Artificial Intelligence Conference in Shanghai. The company says its five-billion-parameter model can generate interactive 720p video at about 20 frames per second on one GPU. That claim places efficiency, rather than visual spectacle alone, at the center of the release. Matrix-Game 3.5 arrived alongside the Mureka V9.5 music generator and Mureka O3 music reasoning model. Kunlun T

OpenAI launched ChatGPT Work on July 9, turning ChatGPT from a conversational assistant into an agent expected to produce finished work. The first ChatGPT Work features cover websites, email workflows, document analysis, spreadsheets, presentations, reports, and scheduled tasks. The change is larger than another model upgrade. OpenAI is separating quick conversations from longer jobs that involve multiple sources, applications, decisions, and deliverables. Work can continue f

Explore the top anythingllm alternatives for 2026. These include Remio, LM Studio, LibreChat, Open WebUI, Nut Studio, Ollama, GPT4All, NotebookLM, LLMChat, and LaunchDarkly. These ai tools give you many choices. Some are open-source. Some are for businesses. Some are easy for beginners. You can find chat platforms and local deployment options. There are also advanced ai features for personal, business, or research use. The global ai market is growing fast. By 2026, experts th

If you want the best ChatGPT alternatives for 2026. You care about features, privacy, and cost. Many users say ChatGPT can be unpredictable, making it hard to use every day. Some people seek tools that work better with their data, while others desire easy ways to connect with other apps. Additionally, some users want more coding options, and others are looking for strong visualization or teamwork features. The table below illustrates why users switch from ChatGPT to other pla

Many people switch from Obsidian to other note-taking apps. Obsidian can be hard to use sometimes. Copying rich text into Obsidian can mess up the format. You may need plugins for simple features. This can make note-taking feel tricky. Some people feel confused because obsidian does not help you organize notes. It takes more work to link your notes together. If you want a smarter note-taking app, you can try these top 10 Obsidian alternatives for 2026: remio, Notion, Evernote

Meeting recording tools with AI features help people and teams work better. In 2026, 55% of users record almost every meeting. 85% say their follow-up is quick and easy. Statistic Value Percentage of users recording close to all meetings 55% Users describing follow-up processes as efficient 85% Users wanting more AI-powered software 46% Users willing to switch for better features 79% SaaS users seeing rapid growth in adoption 50% Top meeting recording software like remio, Ott

Here are the top 10 digital notebook apps you should try in 2026: remio, Notion, Microsoft OneNote, Evernote, GoodNotes, Apple Notes, Google Keep, Zoho Notebook, Obsidian, and Simplenote. Each digital notebook is designed to help you capture ideas, keep your notebooks organized, and find information quickly. The digital notebook market is projected to be worth over $1 billion in 2025 and will continue to grow over the next decade. Research shows that digital notebook apps wit