← Back to feed

NVIDIA Releases Nemotron 3 Super: A 120B Open Model with 5x Higher Throughput for Agentic AI

L4 · DeveloperModels & ReleasesAlphaSignal· 3/12/2026

Developers building production agentic systems will find the throughput gains, open training methodology, and multi-token prediction capabilities directly applicable to reducing inference costs and improving agent responsiveness.

AI Summary

NVIDIA released Nemotron 3 Super, a 120B open-weight model with only 12B active parameters via mixture-of-experts architecture, delivering 5× higher throughput and a 1M . The company is publishing full weights, data, and recipes, positioning it as optimized for building specialized AI agents on Blackwell GPUs.

Excerpt

Nemotron 3 Super is a 120B model with weights, training data, and recipes fully public. It has a 5× higher throughput and 1M context window.

Read Original
0 upvotes · 0 downvotes · 1 min read

Related Articles

L4 · DeveloperModels & ReleasesHacker News
Claude Fable 5.1 and Claude Mythos 5.1

Anthropic launches Claude Fable 5.1 and Mythos 5.1 with 25-45% cost reductions, improved safeguards, and enterprise data privacy controls.

L3 · BuilderModels & ReleasesLatent Space
[AINews] Claude Fable/Mythos 5.1: new SOTA model, 75% cache price cut but 70% more output tokens

Anthropic launched Claude Fable 5.1 and Mythos 5.1 with SOTA performance, 75% cache price reduction but 70% more output tokens per task.

L3 · BuilderModels & Releases@OpenAI
OpenAI (@OpenAI): As we prepare to release Astra, we’re focused on making increasingly capable AI safe and broadly accessible. Astra represents a significant advance in cybersecurity capability, reaching the Critical

OpenAI's upcoming Astra model represents a major advance in cybersecurity capability, reaching the 'Critical' threshold under its Preparedness Framework while prioritizing safety and broad accessibility.

L4 · DeveloperModels & Releases@emollick.bsky.social
Ethan Mollick (@emollick.bsky.social): I think the release of GLM-5.3, a very good open weights models with considerable offensive cyber capability, is worth noticing. Open weights models have effectively no guardrails, and they only provi

GLM-5.3 has been released with significant offensive cyber capabilities and minimal safety guardrails, raising concerns about open-weights model risks.

L1 · CuriousModels & ReleasesOpenAI Blog
Path to Astra: critical capabilities and frontier safeguards

OpenAI announces Astra, its first model to meet the critical cybersecurity capability threshold under its Preparedness Framework with enhanced safeguards.