← all stories other 2 sources · 47m ago

NVIDIA Announces Nemotron 3 Ultra Open Model and Vera Rubin Mass Production

Nemotron 3 Ultra positions NVIDIA as a contender in the open model space against rapidly improving Chinese models, but benchmark results show it trails the top Chinese open model in raw intelligence while leading in throughput.

Reporting from 2 sources: ASCII.jp, GIGAZINE.

NVIDIA Announces Nemotron 3 Ultra Open Model and Vera Rubin Mass Production

NVIDIA CEO Jensen Huang announced the open AI model Nemotron 3 Ultra and the start of mass production of the Vera Rubin AI server during the GTC Taipei 2026 keynote on June 1. Nemotron 3 Ultra is a 550-billion-parameter open model that NVIDIA claims is the highest-performing open model from an American company. Benchmark comparisons against Chinese open models GLM 5.1, Kimi K2.6, and Qwen3.5 show Nemotron 3 Ultra winning in multiple tests and offering superior cost efficiency. However, on the Artificial Analysis Intelligence Index, Nemotron 3 Ultra scored 48 points, below Kimi K2.6's 54 points. NVIDIA emphasized Nemotron 3 Ultra's processing speed advantage, showing it can output more tokens per second than models in its score range. The model is scheduled for release within the first week of June 2026. The Vera Rubin server combines the Rubin GPU and Vera CPU with dedicated storage and networking, targeting agentic AI workloads. Huang also announced the RTX Spark SoC for notebook PCs and the DGX Station desktop PC.

NVIDIA CEO Jensen Huang made the announcements during the GTC Taipei 2026 keynote on June 1. The event was held at 12:00 JST. Alongside Nemotron 3 Ultra and Vera Rubin, Huang announced the RTX Spark SoC for notebook PCs and the DGX Station desktop PC.

The RTX Spark combines an Arm-based CPU, an NVIDIA Blackwell RTX GPU, and up to 128GB of unified memory with full CUDA support. It dynamically allocates memory to improve efficiency for heavy workloads such as simultaneous AI generation, 3D rendering, and multi-model workflows. The chip can run AI models with up to 120 billion parameters locally.

Microsoft announced on May 31 the Surface Laptop Ultra, a notebook PC that adopts the RTX Spark. The Surface Laptop Ultra targets AI developers and creators and is scheduled for release in the latter half of 2026. It has a 15-inch touch-enabled mini LED display, a haptic feedback touchpad, and ports including HDMI, USB-C, USB-A, and a full-size SD card slot. Colors are Platinum and Nightfall. Pricing and detailed specifications will be announced later.

The DGX Station is a high-performance Windows desktop PC with a built-in GB300 chip and up to 748GB of memory. NVIDIA said it can run AI models with up to 1 trillion parameters.

On the Artificial Analysis Intelligence Index, Nemotron 3 Ultra scored 48 points, surpassing Gemma 4 31B's 39 points. The Chinese open model Kimi K2.6 scored 54 points. NVIDIA said Nemotron 3 Ultra is "the highest-performing among American open models, but not as good as advanced Chinese open models."

Synthesized by Yomimono from the 2 cited sources below, including Japanese-language reporting where cited, then editorially reviewed before publishing.

Sources