31 results for “DeepSeek”
NVIDIA submitted its first MLPerf Inference v6.1 preview results for Vera Rubin NVL72, showing up to 3.7x higher throughput than GB300 NVL72 on Qwen3-VL. The numbers signal a leap for rack-scale inference on MoE models.
NVIDIA’s DSX bet is tokens per megawatt: Lambda reports 24% more HGX B200 throughput at 85% power. Eos + Emerald cut 4 MW to 3 MW for the grid — Flex-pattern proof, not a shipped DSX Flex site.
Tencent's Hunyuan team released Hy-MT2 on 21 May 2026 — a new translation model family with instruction-following templates, three sizes up to a 30B-A3B MoE variant, and open weights plus IFMTBench. This is the next generation after the April Hy-MT1.5 phone quantizations.
Anthropic released its September 2026 threat intelligence report on Claude misuse, naming seven harm categories. It alleges three Chinese AI labs relayed user traffic to Claude for model distillation.
Y Combinator CEO Garry Tan urged U.S. regulators to allow domestic open-weight labs to distill American frontier models. He warned that concentrating AI power in one proprietary provider is the true doomer scenario.
After a summer of agent boundary escapes, NVIDIA maps five stack layers and argues security must live in OpenShell-style runtimes—not in harness code models can rewrite.
On September 8, 2026 the PyTorch Foundation said Alibaba Cloud and Cambricon joined as Platinum members and Ant Group as Gold at PyTorch Conference China in Shanghai. The move widens Chinese cloud, chip, and agent-infra participation in vendor-neutral PyTorch governance.
Two dozen companies and organizations have signed an open letter urging United States policymakers to protect open-weight artificial intelligence models. The letter carves out space for one technique that has become contentious in AI.
AI model costs are forcing OpenAI, Meta, xAI and other major labs to shift their sales pitch from raw capability toward cheaper intelligence per task. Enterprise AI bills have become a board-level concern.
NVIDIA detailed how Spectrum-X Ethernet uses hardware-accelerated adaptive routing, congestion control, and plane load balancing to solve traditional Ethernet's failures at giga-scale AI training. The platform delivers 1.6x performance over off-the-shelf Ethernet, near-perfect multi-tenant isolation, and 2.68 ms failover versus 1.08 seconds for standard Ethernet.
Chinese technology and e-commerce giant Alibaba has unveiled Qwen3.8-Max, its largest and most capable artificial intelligence model to date, positioning. The back-to-back releases from Moonshot AI and Alibaba have rattled competitive.
Tencent open-sourced Hy4 preview, a 770B-parameter MoE model with 49B activated weights and a 1M-token context window. The Apache 2.0 release puts a frontier-class model within reach of self-hosted inference stacks.