Description
Qwen's most powerful code model, featuring 480B total parameters with 35B activated through Mixture of Experts (MoE) architecture.
Stats
9.9K Downloads
19 stars
Capabilities
Trained for tool use
Minimum system memory
250GB
Tags
480B
qwen3_moe
Last updated
Updated on July 31byREADME
Qwen3 Coder 480B
Qwen's most powerful code model, featuring 480B total parameters with 35B activated through Mixture of Experts (MoE) architecture.
Key Features:
- Agentic Coding: Comparable performance to Claude Sonnet 4 on coding tasks
- Repository-Scale Understanding: Optimized for large codebases and complex projects
Technical Specifications:
- 480B total parameters, 35B activated (MoE with 160 experts, 8 active)
- 62 layers with Grouped Query Attention (96 Q heads, 8 KV heads)
- Native 262,144 token context length
Note: This model operates in non-thinking mode only and does not generate <think></think> blocks.
Parameters
Custom configuration options included with this model
Repeat Penalty
1.05
Temperature
0.7
Top K Sampling
20
Top P Sampling
0.8
Sources
The underlying model files this model uses