71 results for “storage”
Nvidia has formally opened the cuFile APIs and the vertical storage stack beneath them, moving the interface that lets GPUs read and write storage. Opening cuFile is a strategic departure; this layer has historically lived inside CUDA.
NVIDIA published Vera CPU benchmarks showing up to 3.67x faster CRC32C integrity checks and 3.26x faster Reed-Solomon recovery versus an unnamed x86 CPU. The gains target the storage processing bottleneck that can throttle agentic AI workloads.
Anthropic announced Enterprise Frontier Safeguards, allowing regulated enterprises to store 30-day safety monitoring logs in their own AWS, Azure, or GCP accounts under customer-managed encryption keys with zero vendor human review.
NVIDIA launched DSX Ready, a qualification program for power and cooling products in AI factories. The first categories are battery storage and cooling distribution units, giving builders a vetted partner shortlist.
Cloudflare revealed Cache Transcoding, an internal system built in Pingora that re-encodes cacheable assets into Zstandard before writing to disk, compressing assets to roughly one-third their size and expanding edge storage capacity.
NVIDIA declared Scale-In the fifth pillar of AI networking infrastructure, powered by BlueField-4 DPUs, DOCA software, and Spectrum-X Ethernet. The architecture offloads security, storage, and telemetry from host CPUs at up to 800 Gb/s to keep pace with agentic AI factories.
Anthropic made Computer Use, Skills API, and Files API generally available on the Claude Platform with a new Browser Use tool for web apps. The release adds multi-action turns, HIPAA eligibility, and 1 TB org storage for production agent workloads.
OpenAI says an internal model produced a Navier–Stokes Millennium Prize proof and Lean formalization — ~10,000 agents over 88 hours. It is not claiming the Clay prize.
A research-stage arXiv paper tests whether the factors LLMs cite for their decisions match measured behavioural influence. Across eight Claude, GPT, and Gemini models, cited top-three factors often missed the strongest drivers.
Alibaba's Qwen team open-sourced Qwen-Image-2.1, a 7B model unifying image generation, editing, and transparent RGBA output. Day-0 support covers Diffusers, ComfyUI, vLLM-Omni, and SGLang.
OpenAI upgraded GPT-6 prompt caching with a 30-minute reuse window and new dashboard tools to monitor hit rates. The changes target persistent agent workloads and can cut cached input token costs by up to 90%.
Anthropic launched Claude Opus 5.5, saying it matches Fable 5.1-level performance at 40% lower cost than Opus 5. The model leads on agentic coding benchmarks and adds safeguards for biology and cybersecurity work.