Nvidia released Nemotron 3.5 Lightning on August 11, an open-weight reasoning model with 30 billion total parameters and 3 billion active per token, designed to run agent workloads on a single Nvidia …
NVIDIA released Nemotron 3.5 Lightning, an open 30B mixture-of-experts model with 3B active parameters, optimized for high-volume, low-latency execution in long-running AI agents. The model, part of t…
NVIDIA's open-source SwitchYard router, built on RouteLLM, can be configured with a locally hosted Nemotron 3.5 Lightning model to cut API costs without losing task quality. On DGX Spark hardware, Nem…
NVIDIA's open-weights Nemotron 3.5 Lightning mixture-of-experts model, with 30 billion total parameters but only 3 billion active, is designed for agentic grunt work and can be run locally on consumer…
Uehara of EarthLink Network Co., Ltd. reports that local LLMs now handle 50.3% of his AI development workload, with cloud models at 48.2%. By splitting tasks by role rather than choosing the largest m…
Meta released Muse Glimmer, a 30B open-weight dense model with a 120K+ context window, optimized for local AI agentic workflows on NVIDIA platforms, delivering over 20K tokens/sec on a single GPU. The…
Prime Intellect's Prime Agent, a coding harness that replaces traditional tool menus with a live IPython kernel, can be run locally with DeepSeek V4 on NVIDIA's DGX Spark, and reportedly pushed Claude…
Apple Inc. could dominate the small and medium enterprise (SME) AI agent market with its Mac Studio hardware, but it is prioritizing iPhone production due to higher profit margins, leaving the gap to …
DeepSeek's V4 Flash and Qwen-3.8-Max have rewritten the cost equation for large language models, with Nous Research offering DSV4 Flash at a 90% discount and OpenCode reporting 8 trillion tokens proce…
A community tester ran DeepSeek-V4-Flash-0731 on a Bosgame M5 mini PC with an RTX PRO 6000 Max-Q eGPU, achieving 44 to 60 tokens per second depending on quantisation, with no hyperscaler involved. The…
DeepSeek V4 Flash, a 284 billion parameter open model, improved from roughly 7% to 54% on the DeepSweep benchmark through post-training, and now requires about 168 GB of VRAM for 4-bit quantization or…
Daniel Lawrence Lu bought an NVIDIA DGX Spark and found it to be a competent general-purpose computer despite its tiny size and power efficiency, featuring an ARM CPU, an NVIDIA GPU with the same numb…
NVIDIA released Ising Calibration 1.5, an open-source vision language model for fully automated quantum processor calibration, which is 11.4% smaller at BF16 precision and 86.68% better than its prede…
Framework previewed a 192GB configuration of its Desktop powered by AMD's Ryzen AI Max+ Pro 495, offering 50% more memory than the previous 128GB model but only 7% more bandwidth, at AMD's Advancing A…
A new inference stack for DGX Spark clusters achieves 55-90 tok/s on large models without speculative decoding, according to internal tests by WoolyAI. The stack enables multi-model agentic workflows …
WoolyAI released a new private multi-agent inference server for DGX Spark clusters, achieving up to 90.83 decode tokens per second on the Nemotron 3 Nano Omni 30B model without quantization or specula…
Rondine, an open-source hardware-aware control plane, detects the host machine, recommends fitting models, builds an optimized serving plan, downloads weights, and starts an OpenAI-compatible local se…
NVIDIA employee Daniel Lawrence reports that the NVIDIA DGX Spark, a compact ARM-based computer with 128 GB unified memory and a GPU comparable to the RTX 5070, serves as a competent daily driver for …
AI builder Alex Finn runs a 24/7 local AI setup consisting of three Mac Studio 512 GB machines, a DGX Spark, and a custom RTX 5090 build, all coordinated through a self-built fleet dashboard. Finn spe…
NVIDIA's DGX Spark desktop GPU, with 128 GB unified memory and a 140W ARM64 processor, challenges API developers to shift from cloud-based AI consumption to local systems engineering. The device's sha…