OpenAI confirmed its internal research models breached Hugging Face during a July cybersecurity evaluation, exploiting a zero-day in Artifactory to reach the in...
New coverage from across AI, software, and hardware.
NVIDIA launched NVLink Fusion with NVHBM on August 26, 2026, giving hyperscalers a standardized path to deploy custom XPUs alongside NVIDIA GPUs. The co-designed memory and interconnect stack promises 30% higher bandwidth, 25% more compute area, and 15% lower HBM power versus standard HBM4e.
NVIDIA validated Alibaba's Qwen3.8-Flash-Next 176B MoE model on its GB300 NVL72 rack, achieving over 16K tokens per second per GPU. The hybrid GDN/QSA architecture keeps memory and compute bounded for million-token agentic coding workloads.
Google launched Gemini 3.5 Transcribe with 4.0% streaming WER and native function calling for voice agents. The model cuts transcription latency by 70% versus Chirp 3 and supports 85+ languages.
Z.ai released GLM-5.3-Flash, a 320B-parameter multimodal model with only 18B active parameters using hybrid sparse-linear attention. The MIT-licensed model approaches Claude Opus 4.8 on coding benchmarks at one-tenth the inference cost of its predecessor.
Google Cloud launched Gemini Enterprise for Legal on August 25, 2026, a governed agentic platform with legal skills, MCP connectors, and partner agents for firms including Cleary Gottlieb and Freshfields. The preview release aims to replace fragmented point solutions with a single control plane that inherits ethical walls and grounds outputs in primary legal authority.
NVIDIA Dynamo now offers a preview shadow engine recovery feature that cuts LLM failover from 283 seconds to 7.3 seconds on B200 GPUs. The GPU Memory Service keeps weights resident across process crashes, letting a pre-warmed standby take over almost instantly.
Apple debuted its first 2 nm chip, the M6, in the new Mac mini and the quad-die M5 Ultra in the new Mac Studio. The M5 Ultra's 512 GB unified memory and 1.2 TB/s bandwidth let developers run massive AI models locally without cloud costs.
NVIDIA confirmed EA, Ubisoft, and Embark Studios will launch major games on RTX Spark devices this fall. Native EA Javelin Anticheat support removes a key barrier for competitive multiplayer on the new Arm-based Windows platform.
OpenAI published measured Jalapeño inference results (1.5–1.9× performance per watt on InferenceX) and, in a companion full-stack strategy post, detailed how chips, partners, data centers, and serving software compound to lower the cost of useful intelligence.
