Search “NVIDIA”

al Large 4 (codename Le Chonk), which has 1 trillion parameters with 52 billion active parameters and native multimodal support. ML4 was trained in a European data center on 3,800 Nvidia Grace Blackwell GPUs, improving from 28% to 41% on Terminal-Bench 4.0, tying GLM-5.3 for the best coding-agent score while using an average of 47% fewer tokens per task. Open weights mean enterprises can deploy it on their own hardware and behind their firewalls, and the model offers three reasoning modes: low,

Matched fieldsSummary、Body

continued reinforcement learning over three weeks, raising the Terminal-Bench 4.0 score from 28% to 41%; the Artificial Analysis coding index is 48%, on par with GLM-5.3 while using an average of 47% fewer tokens per task. API pricing drops to $0.70 per million input tokens and $2.10 per million output tokens; the new Mistral Ecosystem Open-Weight License permits commercial use, but reselling model services requires a separate agreement, and the full version requires 8 Nvidia B200 or B300 GPUs.

Matched fieldsSummary、Body

-site information when collective communication hangs, and can be reverted via TORCH DIST USE NCCL2=0 or nccl-legacy. This version also adds float8 e4m3fn and scaled mm support for Apple Silicon, and the new grouped mm improves MoE decoding throughput by 1.32–1.34x on M3 Ultra. The default wheel is upgraded to CUDA 13.2, while a CUDA 13.4 wheel is also provided with experimental support for NVIDIA Rubin GPUs; a ROCm 10.0 wheel is newly added, and Python 3.10 and macOS 14 are no longer supported.

Matched fieldsSummary、Body

AnySpeechRecognizer supports engine switching on top of the Apple SpeechAnalyzer API, with options including NVIDIA Parakeet TDT 0.6B v3 (supporting 25 European languages), Moondream post-trained Parakeet Ultra, the English version Parakeet Unified, and Alibaba Qwen3-ASR (0.6B/1.7B, covering 30 languages and 22 Chinese dialects). These models have been converted to Core ML and published on the Hugging Face Hub; Parakeet takes about 20–40 seconds to transcribe 1 hour of audio, while Qwen3-ASR is

Matched fieldsSummary、Body

After a three-week preview, Mistral AI has released the weights of Mistral Large 4 on Hugging Face. This multimodal model was trained on 3,800 Nvidia accelerators and has a parameter count of one trillion, and it is currently still difficult to run on a gaming PC or a single GPU. Mistral says it leads GLM-5.3 and Kimi K3 on the cybersecurity benchmark CyberGym-E2E, and is on par with GLM-5.3 on the agentic coding benchmark Terminal Bench 4.0, but there is no independent evaluation yet; preview p

Matched fieldsSummary、Body

LC image ray:train-ml-cuda-v1.1, which comes pre-integrated with Ray 2.58.0, PyTorch 2.13.0, CUDA 13.0.2, NCCL 2.29.7, and EFA along with the AWS OFI NCCL plugin, enabling distributed training on EKS without the need to build images yourself. The example uses PyTorch FSDP to fine-tune Qwen2.5-1.5B on 8 NVIDIA L4 GPUs across two g6.12xlarge instances, with cross-node communication going directly over EFA. The image is publicly hosted on Amazon ECR Public, so nodes can pull it without credentials.

Matched fieldsSummary、Body

Matched fieldsSummary、Body

racle announced that it will deploy Google TPU on OCI, along with inference chips from OpenAI, Cerebras, SambaNova, and Positron AI, giving customers more computing options beyond Nvidia GPUs. Oracle says it will become the first hyperscale cloud provider other than Google Cloud to offer TPU; TPU and the Cerebras wafer-scale engine are planned to launch in 2027, and Positron Atlas is already available. At the same time, Oracle is expanding its partnership with Nvidia, with the Vera Rubin NVL72 s

Matched fieldsSummary、Body

been released on Hugging Face, and lightweight testing can run on a single 24GB GPU. The model ranks first on olmOCR-Bench, OmniDocBench, and DocParsingBench, achieves a decoding speed of 558 tokens/s on a single NVIDIA RTX PRO 6000 GPU, and parses 100 pages end-to-end in about 101 seconds. Its post-training stage employs a recursive self-improvement (RSI) loop, in which a general-purpose VLM judges parsing results and generates newly annotated pages, eliminating the need for manual annotation.

Matched fieldsSummary、Body

work as usual: in the Gemma 4 E4B robot dog test, text on a poster was treated as an instruction and changed its movement; in a hospital service robot simulation using Nemotron + NVIDIA Jetson AGX Orin, constructed audio inaudible to the human ear changed the simulated behavior. Researchers also injected messages into a ROS 2/DDS robot that was supposed to remain stationary during a robot bug bounty event, causing it to move. VicOne recommends testing safety limits with adversarial scenarios be

Matched fieldsSummary、Body

its answers are based. When the app starts, it detects the computer's memory and downloads a matching model, ranging from IBM Granite 4.2 8B on devices with 16–31 GB of memory to NVIDIA Nemotron 3.5 Super on Macs with 128 GB or more; the reference library is about 9 GB and contains roughly 930,000 Wikipedia articles, FEMA and CDC guidelines, military medical manuals, and iFixit repair guides. In the author's blind test of 22 power-outage and first-aid questions, Nemotron 3.5 Super and Qwen3.8 2

Matched fieldsSummary

hts of Mistral Large 4 (codename le Chonk). This is a natively multimodal model with 1 trillion total parameters and 49 billion activated parameters, trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own data center in Europe. The company says it scores 48% on the Artificial Analysis Coding Agent Index, roughly on par with GLM-5.3, but consumes 47% fewer Tokens per task on average. The model is released under a modified MIT license: personal, academic, fine-tuning, and conventiona

Matched fieldsSummary、Body

or download after ending its public preview. The model supports text and image processing, with a focus on strengthening defensive cybersecurity, agentic capabilities, and geospatial and grounding capabilities, and is said to achieve state-of-the-art results on agentic benchmarks such as SWE-bench and Terminal Bench. The model was trained on more than 160 languages and is available via Mistral Studio, the Mistral API, OpenRouter, Hugging Face open weights, Vibe Code, and Nvidia-hosted endpoints.

Matched fieldsSummary、Body

s and the gigawatt-scale Prometheus cluster, including its first liquid-cooled network rack IcePack and the 102.4T switch IceCube (128×800G, available with Broadcom Tomahawk 6C or NVIDIA Spectrum-6). For scale-out, the Flattened Planar Fabric connects each GPU to eight network planes, simplifying the Clos architecture from 3 tiers to 2; for scale-across, it introduced the petabit-class super backbone BAG connecting buildings, campuses, and regions. Meta also introduced Tahan, an Ethernet scale-u

Matched fieldsSummary、Body

Musk said on X that SpaceX has secured $40 billion in debt financing to purchase GPUs from suppliers such as Nvidia, and that these GPUs will be delivered in the first and second quarters of 2027. He said SpaceX has an investment-grade credit rating and that Starlink and the launch business generate revenue independent of AI data centers, which is why it was able to obtain this funding; over the past few weeks, the debt market has been nearly closed to companies without independent revenue. Musk

Matched fieldsSummary、Body

ed from scratch purely on human speech, with no text and no pretrained base, supporting full duplex so it can listen and speak at the same time. The model claims inference faster than real time and can run locally on a MacBook Pro M4 Max and above, and the entire model is said to have been trained on a single NVIDIA GB200 NVL72. The team says its goal is to let the model understand tone and emotion, avoiding the safety risk of continuing to respond without perceiving the other person's distress.

Matched fieldsSummary、Body

Schneider Electric has released a new generation of prefabricated power modules and power skids, based on a reference design jointly developed with the Nvidia GB300 NVL72 AI server system, supporting up to 142kW of power density per rack and integrating power management and liquid cooling control. The product uses Galaxy VXL UPS technology and offers three capacities: 2MW, 2.25MW, and 2.5MW, with the 2.5MW being its highest-capacity single-container prefabricated power solution to date. The prod

Matched fieldsSummary

51World's subsidiary 51Sim was selected for the NVIDIA Halos SAFE ecosystem list announced by NVIDIA on October 21, making it the only Chinese company on the list. The company's share price rose 58.92% cumulatively over the week, with a single-day gain of 30.31% on October 23. Halos SAFE is a framework launched by NVIDIA in the autonomous driving field that integrates simulation and real-vehicle testing, covering the full cycle of design, deployment, and validation. This week, southbound funds r

Matched fieldsSummary、Body

itecture, can listen and speak at the same time, and can run locally on a MacBook M4 Max or a more powerful consumer-grade computer. The model was trained entirely on human speech and a single NVIDIA GPU, and the company says expressions such as emotion, pauses, and laughter are built into the model rather than driven by a text pipeline. Dessa is not currently publicly available, and Block is recruiting security researchers to test the risk that it could be used to deceive conversation partners.

Matched fieldsSummary、Body

GLang matches or exceeds DeepInfer on DeepInfer's own benchmarks: up to 1.04x prefill throughput on 4x GB300 and 4x B300, and 1.1%–7.4% shorter decode step time on 64x GB300. SGLang has integrated DeepSeek's Mega series kernels, DSpark speculative decoding, SWA bounded replay, and two Engram placement methods, and has filled in KV cache, prefill/decode disaggregation, and inter-replica load balancing that DeepInfer does not include. SGLang supports NVIDIA GB300, B300, B200, H200, and AMD MI355X.

Matched fieldsSummary、Body