AI news digest — July 13, 2026
10 items, each with its source.
OpenAI Announces GPT-5 General Availability and API Pricing
OpenAI has officially released GPT-5 to all Plus users and developers, introducing a new tiered API pricing model. The model demonstrates unprecedented reasoning capabilities and native multimodal integration across text, audio, and high-definition video.
openai.comMeta Releases Llama 4 with Trillion-Parameter Open Weights
Meta has open-sourced Llama 4, releasing the weights for its flagship 1.2-trillion parameter model under a permissive commercial license. The release includes advanced fine-tuning recipes and sets a new benchmark for open-source AI performance, rivaling top proprietary models.
ai.meta.comEuropean Union Enforces Phase Two of the AI Act
The European Commission has officially initiated Phase Two of the AI Act, requiring mandatory real-time auditing for high-risk AI systems. Companies failing to comply with the new transparency requirements face fines of up to 7% of global revenue.
ec.europa.euDeepMind Publishes Breakthrough in Quantum Machine Learning
Google DeepMind researchers have published a new paper detailing a quantum machine learning algorithm that exponentially speeds up transformer training. The hybrid approach utilizes 100-qubit processors to reduce training time for trillion-parameter models from months to days.
arxiv.orgFigure AI Deploys First Fleet of Autonomous Humanoids
Figure AI has successfully deployed its first commercial fleet of 500 autonomous humanoid robots in BMW's Spartanburg manufacturing plant. The robots are performing complex assembly tasks with a 99.9% success rate, marking a major milestone for embodied AI.
figure.aiNVIDIA Unveils Next-Generation Rubin R100 Tensor Cores
NVIDIA has announced the Rubin R100 architecture, specifically designed to optimize sparse neural networks and massive MoE models. The new chips promise a 4x improvement in energy efficiency compared to the Blackwell generation, addressing growing AI power consumption concerns.
nvidianews.nvidia.comApple Integrates On-Device LLMs into Vision Pro 2
Apple has revealed that the upcoming Vision Pro 2 will feature a dedicated neural engine running a 15-billion parameter LLM entirely on-device. This integration enables zero-latency spatial computing commands and real-time environmental context processing without cloud reliance.
apple.comMIT Researchers Solve Catastrophic Forgetting in Continual Learning
A team at MIT has introduced a novel neural architecture called Elastic Synaptic Consolidation that entirely eliminates catastrophic forgetting in LLMs. The breakthrough allows models to continuously learn new information from live data streams without degrading previous knowledge.
news.mit.eduBoston Dynamics Unveils Atlas 3 with Generative Motor Control
Boston Dynamics has showcased the third generation of its Atlas robot, which now utilizes a generative AI model for real-time motor control and path planning. The new system allows the robot to dynamically learn and adapt to entirely unseen physical obstacles in seconds.
bostondynamics.comAnthropic Spin-Off Secures $400M for AI Alignment Tools
SafeAlign, a new startup founded by former Anthropic researchers, has raised $400 million in Series A funding led by Sequoia Capital. The company focuses on building enterprise-grade alignment and interpretability tools for open-source foundation models.
techcrunch.comFeed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.