Qwen3-30B-A3B (MoE) Architecture Parameters, Benchmarks & VRAM
30.5B total / 3.3B active MoE delivering ultra-high token throughput on 24GB GPUs and Apple Silicon.
🧬 Qwen3-30B-A3B (MoE) Architecture & Tensor Specifications
Verified directly against HuggingFace (Qwen/Qwen3-30B-A3B) config.json & Model Card
1. Backbone & Dimensions
qwen3_moe- Total / Active Params
- 30.5B / 3.3B
- Non-Embedding Params
- 29.9B
- Hidden Size (d_model)
- 2,048
- FFN Intermediate Size
- 6,144
- MoE Expert Dim
- 768
- MoE Experts
- 128 Routed · Top-8
- Activation / Norm
- SiLU (SwiGLU) · ε=1e-6
- Native Tensor Dtype
- BF16
2. Attention & Layer Layout
GQA- Num Hidden Layers
- 48 Layers
- Hidden Layer Layout
- 48-Layer Fine-Grained MoE (128 Experts, 8 Activated)
- Attention Heads (Q / KV)
- 32 Q / 4 KV (8:1 GQA)
- Head Dimension
- 128
3. Context, RoPE & Tokenizer
Tokenizer- Native Context Length
- 32,768 tokens
- Max Extended Context
- 131,072 tokens
- RoPE Configuration
- RoPE (θ=1,000,000 · 原生 32K · YaRN 4x→128K)
- Vocab Size
- 151,936 (Untied)
- Special Token IDs
- BOS: 151643 · EOS: [151645,151643] · PAD: 151643
- Language Support
- 119 languages & dialects
📋 Qwen3-30B-A3B (MoE) Official API & System Card Specifications
Verified against official Alibaba Qwen developer documentation & pricing
1. Identity & Capabilities
API Spec- API Model ID
- Qwen/Qwen3-30B-A3B-Instruct-2507
- Release Date
- 2025-04-29
- Weight License
- Open-Weight (API Spec)
- Native Reasoning
- —
- Function Calling & JSON Schema
- ✅ 原生支持
- Batch API 异步批处理
- ✅ Supported (50% off)
2. Context, Output & Modalities
Limits- Max Context Window
- 128,000 tokens (128K)
- Max Output Tokens
- 262,000 tokens
- Input Modalities
- Text
- Output Modalities
- Text
- Direct China Access
- ✅ Direct CN + RMB
- Deployment Mode
- Open-Weight + Cloud API
3. Official Pricing & Cache
CNY / 1M- Input / 1M Tokens
- ¥ 0.7
- Cached Input / 1M
- ¥ 0.7
- Output / 1M Tokens
- ¥ 2.8
- Tiered Pricing Notes
- SiliconFlow: Open-weight cloud hosting on SiliconFlow
🏆 Official Benchmark Results
Sourced from official System Card / Model Card & technical reports
An improved version of the MMLU benchmark featuring manually re-annotated questions to identify and correct errors in the original dataset. Provides more reliable evaluation metrics for language models by addressing dataset quality issues found in the original MMLU.
C-Eval is a comprehensive Chinese evaluation suite designed to assess advanced knowledge and reasoning abilities of foundation models in a Chinese context.
PhD-level Google-proof science reasoning questions
MATH-500 is a subset of the MATH dataset containing 500 challenging competition mathematics problems from AMC 10, AMC 12, AIME, and other mathematics competitions.
American Invitational Mathematics Examination 2024, consisting of 30 challenging mathematical reasoning problems from AIME I and AIME II competitions.
All 30 problems from the 2025 American Invitational Mathematics Examination (AIME I and AIME II), testing olympiad-level mathematical reasoning with integer answers from 000-999.
LiveCodeBench is a holistic and contamination-free evaluation benchmark for large language models for code.
Berkeley Function Calling Leaderboard v3 (BFCL-v3) is an advanced benchmark that evaluates large language models' function calling capabilities through multi-turn and multi-step interactions.
🎛️ Official Sampling & Reasoning Parameters
Recommended generation settings & API parameter constraints
💡 Do NOT use greedy decoding (temp=0); use 38,912 max tokens for competition math/code.
💡 Recommended for fast general chat; presence_penalty 0~2 reduces repetition.
📊 Qwen3-30B-A3B (MoE) Full Benchmark Scorecard
8 public benchmarks · ranks among 154 models · synced 2026-11-20
💻 Coding 1 items
🤖 Agents & Tools 1 items
📐 Math 3 items
American Invitational Mathematics Examination 2024, consisting of 30 challenging mathematical reasoning problems from AIME I and AIME II competitions.
LiveBench is a challenging, contamination-limited LLM benchmark that addresses test set contamination by releasing new questions monthly based on recently-released datasets, arXiv papers, news articles, and IMDb movie synopses.
All 30 problems from the 2025 American Invitational Mathematics Examination (AIME I and AIME II), testing olympiad-level mathematical reasoning with integer answers from 000-999.
🧠 Reasoning & Science 3 items
Arena-Hard-Auto is an automatic evaluation benchmark for instruction-tuned LLMs consisting of 500 challenging real-world prompts curated by BenchBuilder. It includes open-ended software engineering problems, mathematical questions, and creative writing tasks.
Multi-IF benchmarks LLMs on multi-turn and multilingual instruction following. It expands upon IFEval by incorporating multi-turn sequences and translating English prompts into 7 other languages, resulting in 4,501 multilingual conversations with three turns each.
A challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry. Questions are Google-proof and extremely difficult, with PhD experts reaching 65% accuracy.
Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).
⚡ Qwen3-30B-A3B (MoE) Interactive VRAM & Token Speed Calculator
Physical 32 GB · Load 66.7%
MoE: reads only 2 GB active weights/step
Enable Q8/Q4 KV quantization to scale batch concurrency
OLLAMA_CONTEXT_LENGTH=16384 ollama run qwen3:30b-a3bRTX 5090 Fits Qwen3-30B-A3B (MoE) · Cost & TCO
Approx. $2,199 (~$69/GB VRAM) · 575W TDP
📊 Qwen3-30B-A3B (MoE) Quantization VRAM Breakdown (16K Context)
| Quantization | Bits/Weight | Weights | KV Cache (16K) | Total VRAM | Quality Note |
|---|---|---|---|---|---|
| FP16 / BF16 | 16 bpw | 61 GB | 1.61 GB | 64.28 GB | Lossless Full Precision |
| FP8 / Q8_0 | 8.5 bpw | 32.41 GB | 1.61 GB | 35.11 GB | 99.9% Near-Lossless |
| GGUF Q6_K | 6.56 bpw | 25.01 GB | 1.61 GB | 27.57 GB | Ultra High Quality |
| GGUF Q5_K_M | 5.69 bpw | 21.69 GB | 1.61 GB | 24.19 GB | High Balanced Quality |
| Q4_K_M / AWQ | 4.85 bpw | 18.49 GB | 1.61 GB | 20.92 GB | Sweet Spot (Recommended) |
| GGUF IQ4_XS | 4.25 bpw | 16.2 GB | 1.61 GB | 18.59 GB | Compact 4-bit |
| GGUF Q3_K_M | 3.91 bpw | 14.91 GB | 1.61 GB | 17.27 GB | Moderate Perplexity Loss |
| GGUF Q2_K | 3.15 bpw | 12.01 GB | 1.61 GB | 14.31 GB | Extreme Compression |
🖥️ GPU & Apple Silicon Compatibility for Qwen3-30B-A3B (MoE) (Q4_K_M)
Frequently Asked Questions
What are the exact architecture parameters of Qwen3-30B-A3B (MoE)?
Qwen3-30B-A3B (MoE) uses Qwen3MoeForCausalLM with 48 layers (48-Layer Fine-Grained MoE (128 Experts, 8 Activated)), hidden_size=2048, 32 Q heads / 4 KV heads (head_dim=128), and vocab_size=151,936.
How much VRAM does Qwen3-30B-A3B (MoE) require?
At 4-bit quantization (Q4_K_M, 16K context), Qwen3-30B-A3B (MoE) requires 20.92 GB VRAM. Full FP16 requires 64.28 GB.