Qwen iconQwen: Qwen3 30B A3B

Model Type

Open weight model icon

Open Weight Model

30B parameters

Recommended Use Cases

Text Generation

Try Qwen3 30B A3B

Qwen3 30B A3B is Alibaba's efficient MoE language model with 30B total parameters and 3B active per token, outperforming QwQ-32B while using only 10% of the activated parameters.

The small MoE model, Qwen3-30B-A3B, outcompetes QwQ-32B with 10 times of activated parameters.

  • Qwen Team

Overview

Qwen3-30B-A3B is the smaller MoE model in the Qwen3 series, offering an exceptional balance of capability and efficiency. It uses 128 experts with 8 activated per token, achieving performance comparable to much larger dense models while significantly reducing inference costs.

Key Features

  • MoE efficiency: 30B total, only 3B active per token
  • Hybrid thinking: Toggle between thinking and non-thinking modes
  • 128K context: Native support, extendable to 131K with YaRN
  • Outperforms QwQ-32B: Better reasoning with 10x fewer active parameters
  • 119 languages: Broad multilingual support

Technical Specifications

SpecificationValue
Total Parameters30.5B
Active Parameters3.3B
Experts128 (8 activated per token)
Layers48
ArchitectureMoE transformer
Context Length128K tokens (131K with YaRN)
Training Data36T tokens
Release DateApril 2025
LicenseApache 2.0

When to Use Qwen3-30B-A3B

Choose Qwen3-30B-A3B when you need:

  • Strong performance with efficient inference
  • Reasoning capability at reduced cost
  • Self-hosted deployment with limited resources
  • Balance between capability and speed

Consider alternatives when:

  • Maximum capability → Qwen3-235B-A22B
  • Simpler architecture → Qwen3-32B (dense)
  • Even more efficiency → Qwen3-Next-80B-A3B (better throughput)

Updated Variants

  • Qwen3-30B-A3B-Instruct-2507: 256K context, non-thinking only
  • Qwen3-30B-A3B-Thinking-2507: 256K context, thinking only

Availability

  • Open Weights: Hugging Face (Qwen/Qwen3-30B-A3B)
  • API: OpenRouter, various providers
  • Local: Ollama, LMStudio, vLLM, SGLang

Role in Series

Qwen3 text models by architecture:

Dense models: 0.6B, 1.7B, 4B, 8B, 14B, 32B

MoE models:

  1. Qwen3-30B-A3B: Efficient MoE (this model)
  2. Qwen3-235B-A22B: Flagship MoE

Links

Qwen3 30B A3B | Try That LLM