Chinese AI firm Moonshot AI releases Kimi K2 Thinking, an open-source reasoning model that outperforms GPT-5 and Claude Sonnet 4.5, achieving record 44.9% on Humanity's Last Exam benchmark and marking a potential turning point as Chinese models now rival leading US systems.
Google Research unveils Nested Learning, a revolutionary AI paradigm that solves catastrophic forgetting by treating models as interconnected, multi-level systems where components update at different frequencies, enabling machines to learn new tasks without losing previous knowledge.
Generalist AI releases GEN-θ, breakthrough robot foundation models trained on 270,000 hours of real-world manipulation data, discovering a critical 7B parameter intelligence threshold where robots dramatically improve at complex tasks and need only thousands of training steps to master new skills.
Google Cloud launches revolutionary Ironwood TPUs delivering 10X performance boost over previous generation chips, while AI company Anthropic prepares to harness up to one million TPUs as customers report dramatic improvements in speed and cost efficiency.
NVIDIA releases open-source NeMo Automodel library that delivers breakthrough 280 TFLOPs/sec GPU performance for AI training, enabling developers to train massive mixture-of-experts models in PyTorch with simplified infrastructure while achieving near-linear scaling across thousands of GPUs.
Google unleashes Ironwood, its seventh-generation TPU chip that delivers four times faster AI processing than previous models and connects up to 9,216 chips in massive pods, with Anthropic planning to deploy 1 million units for Claude as the tech giant battles Nvidia's GPU dominance.
Lambda strikes massive multibillion-dollar partnership with Microsoft to construct AI infrastructure featuring tens of thousands of Nvidia chips, serving over 200,000 developers while planning a Kansas City AI factory by 2026.
Chinese researchers at Peking University develop groundbreaking analog chip using resistive memory that delivers 1,000x faster performance than top Nvidia and AMD GPUs while consuming 100x less energy, solving precision issues that have plagued analog computing for decades.
Large reasoning models now demonstrate genuine human-like thinking abilities through chain-of-thought reasoning, using pattern matching, working memory, and backtracking search to outperform average humans on logic tests, proving AI has evolved beyond simple pattern matching to possess real problem-solving capabilities.
University of Surrey scientists develop revolutionary brain-mimicking AI called Topographical Sparse Mapping that connects neurons only to nearby related ones, dramatically slashing energy consumption by over a million kilowatt-hours while boosting performance through biological pruning processes that refine connections during learning.