---
format: "aidr-story-markdown/v1"
id: "60a1c88500e1715f3749263fdc6554913f3dd370a5c3a54d878565a06b85203d"
canonical_url: "https://aidr.today/60a1c885?lang=en"
title: "Prime Intellect Releases Prime Mega MoE: Trainable Fused MoE Kernel, 3.11x Faster Forward Pass"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-29T18:28:57.000Z"
category: "Frameworks"
topics: ["deepseek","llm","framework"]
source_urls: ["https://www.marktechpost.com/2026/10/29/prime-intellect-prime-mega-moe-fused-moe-kernel-training/"]
summary: "Prime Intellect released Prime Mega MoE, a backward pass for DeepSeek's fused Mega MoE kernel that makes the kernel trainable and speeds Qwen3 MoE training steps by up to 34% in prime-rl."
---

# Prime Intellect Releases Prime Mega MoE: Trainable Fused MoE Kernel, 3\.11x Faster Forward Pass

> [Open the canonical story](<https://aidr.today/60a1c885?lang=en>)

**Published:** 2026-10-29T18:28:57.000Z
**Category:** Frameworks
**Topics:** deepseek, llm, framework

## Summary

Prime Intellect released Prime Mega MoE, a backward pass for DeepSeek's fused Mega MoE kernel that makes the kernel trainable and speeds Qwen3 MoE training steps by up to 34% in prime\-rl\.

## Sources

- [Story source](<https://www.marktechpost.com/2026/10/29/prime-intellect-prime-mega-moe-fused-moe-kernel-training/>)