2026
10,721 articles · 136 days · 10 categories · avg 79 per day
Meta launches Muse Spark 1.1 with a public API and better tool calling
Meta has unveiled Muse Spark 1.1, the first version of its AI model with public API access and improved agentic and automation capabilities.
Tencent to buy back AI startup Manus from Meta for $2 billion
China's Tencent is ready to finance a buyback of Singapore-based AI startup Manus from Meta Platforms after Chinese authorities blocked the deal.
Biren Technology, MoKe Shensi and Step Build the First Optoelectronic AI Product
Three Chinese companies, MoKe Shensi, Biren Technology and Step, signed an agreement in Beijing and released the world's first AI compute product built on hybrid…
Lilian Weng on harness engineering and recursive self-improvement: a review of 35 papers
Agent orchestration matters more than model choice, arXiv study finds
Why AI Infrastructure Has to Evolve for Agents — Akshat Bubna, Modal CTO
SageMath lifts LLM agents' math performance by 9.7pp on average
DeepSeek V3.2 reaches 67% accuracy on the ARC-AGI abstract reasoning test
Researchers show the open DeepSeek V3.2 model solving ARC-AGI-1 tasks with 67% accuracy and no special training, using two-stage agentic architectures.
Claude goes into production: UST brings Anthropic's AI to chip validation
Anthropic and UST are integrating Claude Code into the iDEC chip validation platform, cutting test cycles from 4 days to 2 days, with expansion into healthcare…
Neolix pitches autonomous logistics as physical AI, expanding into the Middle East
AgentLens: Code-Agent Evaluation Shifts From Outcomes to Process
Malaysian PM Anwar Ibrahim to launch an AI double with his own voice
MiniMax seeks $2 billion for AI model development through share sale
NVIDIA Nemotron-Labs-3-Puzzle-75B: 37% Smaller Model Doubles Server Throughput
NVIDIA has released a compressed version of its Nemotron-3-Super LLM. Using the Puzzle method, which alternates compression and distillation, the model is 37%…
neueHCT Names Jiang Jingfang President to Drive Expansion in New Markets
On July 10, 2026, neueHCT announced Jiang Jingfang as its new president. She will lead commercialization and global expansion of its autonomous driving solutions.
Google Research releases SensorFM, a model for 35 health prediction tasks
Google Research trained the SensorFM foundation model on 1 trillion minutes of data from 5 million Fitbit and Pixel Watch users to predict health outcomes.
OpenAI and Google sell AI to China via Singapore, sidestepping US sanctions
Self-Reflective Program Search: Apple Improves How LLMs Handle Long Contexts
Facewatch launches police integration: in-store facial recognition alerts in 4 seconds
Apple develops TGPO to teach video models to understand time
Anthropic's secondary-market valuation hits $1.2 trillion, up 550% in a year
Anthropic's valuation on secondary markets has soared to $1.2 trillion, a 550% rise in a year, driven by how scarce its shares are among investors.
NVIDIA NeMo generates synthetic financial data for AI
NVIDIA has unveiled a method for generating synthetic financial data to train LLMs, tackling the imbalance of rare events in real-world news.
New Zealand's First AI Data Center: Datagrid to Invest $2 Billion, but Residents Object
Rongje expects profit to grow roughly tenfold in the first half of 2026
SynthAVE: 21 Models Label 12,000 Products in Place of Humans
DeLS-Spec speeds up LLM inference without retraining models
Remote Sensing Foundation Models: How to Adapt AI to Satellite Data
A study finds that foundation models need dedicated adaptation for satellite data, because the physics of the measurements differs from that of ordinary computer…
Tabular Data and Images: Why the Choice of Encoder Matters
Researchers have for the first time evaluated tabular models as encoders for multimodal learning and found that choosing the right one significantly affects quality.
A video model that respects physics: Chinese team open-sources a foundation for training robots
A Chinese video model tackles a critical problem: AI video often breaks physics, misleading robots that learn real-world actions from it.
Mental Disorders in RL Agents: A Dose-Response Study of Anxiety and Mania
Ant Lingbo launches LingBot-VA 2.0, billed as the first robot action model built natively for the physical world
Synthetic Data for Rare Languages: Boosting Julia Code Generation by 14 Points
