Public story · 2026-03-22 · source-backed

Nemotron Cascade 2: NVIDIA's 30B-A3B MoE Outperforms Qwen3.5-35B

Issue

2026-03-22

Confidence

source-backed

Sources

Redaction

passed

Story

3B active parameters, beats Qwen3.5-35B-A3B on AIME 2025 (92.4 vs 91.9), LiveCodeBench v6 (87.2 vs 74.6), and surpasses the larger Nemotron-3-Super-120B. Available on Ollama and HuggingFace under open license. Source

Ollama supports Qwen / Shared entities / Same source domain / Shared topic / What happened next
Qwen3.8-2.4T-A95B: 2.4 trillion parameters, open weights, day-0 vLLM.
Linked by a graph relationship (Ollama supports Qwen); both cover LocalLLaMA, NVIDIA, Qwen3; reported by the same outlet (reddit.com).
Ollama supports DeepSeek / Shared entities / Same source domain / What happened next
Meta ships Muse Glimmer, a 30B open agent model
Linked by a graph relationship (Ollama supports DeepSeek); both cover LocalLLaMA, NVIDIA, Ollama, Qwen3; reported by the same outlet (reddit.com).
Ollama supports Qwen / Shared entities / Same source domain / What happened next
Google Ships Gemma 4 Under Apache 2.0. Four Models, Arena #3, and the License That Changes Everything.
Linked by a graph relationship (Ollama supports Qwen); both cover HuggingFace, LocalLLaMA, Ollama, Qwen3; reported by the same outlet (reddit.com).
Ollama supports DeepSeek / Shared entities / Same source domain / Shared topic / What happened next
DeepSeek V4 Drops: MIT-Licensed Frontier Model at 1/6th the Cost, Running on Chinese Chips
Linked by a graph relationship (Ollama supports DeepSeek); both cover LiveCodeBench, LocalLLaMA, NVIDIA; reported by the same outlet (reddit.com).
NVIDIA released Nemotron / Shared entities / Shared topic / What happened next
NVIDIA Nemotron 3 Super: 120B/12B Active MoE Open Model for Agentic AI at 5x Prior Throughput
Linked by a graph relationship (NVIDIA released Nemotron); both cover Nemotron, NVIDIA, Super; overlapping topics (active, nemotron, nvidia, parameter).
Ollama supports Qwen / Shared entities / Same source domain / Shared topic / Earlier coverage
Practitioner Benchmark: Nemotron 3 Nano 4B Fails Where Qwen 3.5 4B Passes.
Linked by a graph relationship (Ollama supports Qwen); both cover LocalLLaMA, Nemotron, NVIDIA; reported by the same outlet (reddit.com).
Ollama supports Qwen / Shared entities / Same source domain / What happened next / Tension
A 27B Model Just Matched the 400B One. It Fits on Your Laptop.
Linked by a graph relationship (Ollama supports Qwen); both cover LocalLLaMA, Ollama, Qwen3; reported by the same outlet (reddit.com).
Ollama supports Qwen / Shared entities / Same source domain / What happened next / Downstream implication
llama.cpp Gets Backend-Agnostic Tensor Parallelism, and the NVIDIA Tax Just Got Optional
Linked by a graph relationship (Ollama supports Qwen); both cover LocalLLaMA, NVIDIA, Qwen3; reported by the same outlet (reddit.com).

Source trail

Source

Entities

A3B A3B MoE Outperforms Qwen3 AIME HuggingFace LiveCodeBench LocalLLaMA Nemotron Nemotron Cascade NVIDIA Ollama Qwen3 Super

Provenance

Canonical issue: Ramsay Research Agent — 2026-03-22
AI generated: no
Story unit: 2026-03-22-nemotron-cascade-2-nvidia-s-30b-a3b-moe-outperforms-qwen3-5-35b
Labels: source-backed, canonical briefing excerpt

Fetching from the wire…

Nemotron Cascade 2: NVIDIA's 30B-A3B MoE Outperforms Qwen3.5-35B

Story

Related stories

Qwen3.8-2.4T-A95B: 2.4 trillion parameters, open weights, day-0 vLLM.

Meta ships Muse Glimmer, a 30B open agent model

Google Ships Gemma 4 Under Apache 2.0. Four Models, Arena #3, and the License That Changes Everything.

DeepSeek V4 Drops: MIT-Licensed Frontier Model at 1/6th the Cost, Running on Chinese Chips

NVIDIA Nemotron 3 Super: 120B/12B Active MoE Open Model for Agentic AI at 5x Prior Throughput

Practitioner Benchmark: Nemotron 3 Nano 4B Fails Where Qwen 3.5 4B Passes.

A 27B Model Just Matched the 400B One. It Fits on Your Laptop.

llama.cpp Gets Backend-Agnostic Tensor Parallelism, and the NVIDIA Tax Just Got Optional

Source trail

Entities

Provenance

Nemotron Cascade 2: NVIDIA's 30B-A3B MoE Outperforms Qwen3.5-35B

Story

Related stories

Qwen3.8-2.4T-A95B: 2.4 trillion parameters, open weights, day-0 vLLM.

Meta ships Muse Glimmer, a 30B open agent model

Google Ships Gemma 4 Under Apache 2.0. Four Models, Arena #3, and the License That Changes Everything.

DeepSeek V4 Drops: MIT-Licensed Frontier Model at 1/6th the Cost, Running on Chinese Chips

NVIDIA Nemotron 3 Super: 120B/12B Active MoE Open Model for Agentic AI at 5x Prior Throughput

Practitioner Benchmark: Nemotron 3 Nano 4B Fails Where Qwen 3.5 4B Passes.

A 27B Model Just Matched the 400B One. It Fits on Your Laptop.

llama.cpp Gets Backend-Agnostic Tensor Parallelism, and the NVIDIA Tax Just Got Optional

Source trail

Entities

Provenance