RUN THIS LLM

Search local LLM hardware requirements

← Back to all models

ERNIE 4.5 21B MoE Thinking

Baidu · 21B MoE · Reasoning

Baidu's lightweight thinking MoE: 21B total, 3B active per token, with 128K context and function calling. Post-trained for deeper reasoning on math, science, and code; Apache 2.0.

VRAM Requirements

QuantizationVRAM
Q4_K_M (smallest)12.6 GB
Q8_0 (balanced)23.1 GB
FP16 (full quality)42 GB

Specifications

Loading interactive analysis...

RTL
Add Run This LLM to your home screen
Get the Chrome Extension
Look up hardware requirements for any model right from your browser sidebar — no tab switching needed.
Add to Chrome — It's Free