Moe

4 articles tagged with "Moe"

Visual representation of Qwen3-Coder-Next's sparse architecture, showing 3B active parameters against 80B total

Alibaba Releases Qwen3-Coder-Next, Efficient Coding Model

Alibaba's Qwen3-Coder-Next activates just 3B of 80B parameters, hitting 70.6% on SWE-Bench while running on consumer hardware.

Andrés MartínezFeb 4, 20262 min

$Abstract visualization of sparse neural network activation showing a small fraction of illuminated nodes against a dark background, representing efficient parameter usage in AI models.$

LLMs & Foundation Models

StepFun's Step-3.5-Flash Matches Frontier Models at a Fraction of the Size

Shanghai startup's new open-source model claims benchmark wins over DeepSeek V3.2 with dramatically smaller active parameter count

Andrés MartínezFeb 2, 20264 min

Image Generation

Tencent Releases HunyuanImage 3.0-Instruct for Image Editing

The Instruct version brings Chain-of-Thought processing to the 80B-parameter open-source model.

Andrés MartínezJan 26, 20262 min

Visualization of sparse neural network activation patterns representing ERNIE 5.0's Mixture-of-Experts architecture

LLMs & Foundation Models

Baidu Officially Launches ERNIE 5.0, Its 2.4 Trillion Parameter Omni-Modal Model

The model hits #8 globally on LMArena and #2 in math reasoning, though Western labs have already moved on to newer models.

Oliver SentiJan 24, 20264 min

Stay Ahead of the AI Curve

Get the latest AI news, reviews, and deals delivered straight to your inbox. Join 100,000+ AI enthusiasts.

By subscribing, you agree to our Privacy Policy. Unsubscribe anytime.