---
format: "aidr-story-markdown/v1"
id: "e2a9d09fb5d27111288ece149d3f916ba4773e2166449c254e4b15958f1d0ae4"
canonical_url: "https://aidr.today/e2a9d09f?lang=en"
title: "Rigel AI Hits Llama 3.2 Performance Using Under 1% of FLOPs"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-23T05:44:58.000Z"
category: "Research"
topics: ["llm","mamba","inference","efficiency","research","infra","architecture"]
source_urls: ["https://marketbrief.now/ai/rigel-ai-hits-llama-32-performance-using-under-1percent-of-flops-ada6dfb2","https://huggingnews.com/ai/rigel-ai-hits-llama-32-performance-using-under-1percent-of-flops-ada6dfb2"]
summary: "A new hybrid Mamba-2 machine learning architecture operates with 360 million active parameters to achieve high computational efficiency. The 2.3 billion parame…"
---

# Rigel AI Hits Llama 3\.2 Performance Using Under 1% of FLOPs

> [Open the canonical story](<https://aidr.today/e2a9d09f?lang=en>)

**Published:** 2026-09-23T05:44:58.000Z
**Category:** Research
**Topics:** llm, mamba, inference, efficiency, research, infra, architecture

## Summary

A new hybrid Mamba\-2 machine learning architecture operates with 360 million active parameters to achieve high computational efficiency\. The 2\.3 billion parame…

## Sources

- [Story source](<https://marketbrief.now/ai/rigel-ai-hits-llama-32-performance-using-under-1percent-of-flops-ada6dfb2>)
- [Story source](<https://huggingnews.com/ai/rigel-ai-hits-llama-32-performance-using-under-1percent-of-flops-ada6dfb2>)