Qwen3-32B Architecture Parameters, Benchmarks & VRAM
Classic dense 32.8B Qwen3 model with hybrid thinking mode; fits inside 24GB GPUs at Q4_K_M (8K context).
🧬 Qwen3-32B Architecture & Tensor Specifications
Verified directly against HuggingFace (Qwen/Qwen3-32B) config.json & Model Card
1. Backbone & Dimensions
qwen3- Total / Active Params
- 32.8B (Dense)
- Non-Embedding Params
- 31.2B
- Hidden Size (d_model)
- 5,120
- FFN Intermediate Size
- 25,600
- Activation / Norm
- SiLU (SwiGLU) · ε=1e-6
- Native Tensor Dtype
- BF16
2. Attention & Layer Layout
GQA- Num Hidden Layers
- 64 Layers
- Hidden Layer Layout
- 64-Layer Dense Transformer (GQA + SwiGLU + QK-Norm)
- Attention Heads (Q / KV)
- 64 Q / 8 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
📋 Qwen3-32B Official API & System Card Specifications
Verified against official Alibaba Qwen developer documentation & pricing
1. Identity & Capabilities
API Spec- API Model ID
- Qwen/Qwen3-32B
- Release Date
- 2025-04-29
- Weight License
- Open-Weight (API Spec)
- Native Reasoning
- ✅ Supported
- 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
- 131,000 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
- ¥ 1
- Cached Input / 1M
- ¥ 1
- Output / 1M Tokens
- ¥ 4
- 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.
📊 Qwen3-32B Full Benchmark Scorecard
8 public benchmarks · ranks among 154 models · synced 2026-11-20
💻 Coding 2 items
LiveCodeBench is a holistic and contamination-free evaluation benchmark for large language models for code.
Aider is a comprehensive code editing benchmark based on 133 practice exercises from Exercism's Python repository, designed to evaluate AI models' ability to translate natural language coding requests into executable code that passes unit tests.
🤖 Agents & Tools 1 items
📐 Math 4 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.
A competitive programming benchmark using problems from the CodeForces platform. The benchmark evaluates code generation capabilities of LLMs on algorithmic problems with difficulty ratings ranging from 800 to 2400.
Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).
⚡ Qwen3-32B Interactive VRAM & Token Speed Calculator
Physical 32 GB · Load 79.8%
Actual speed varies ~±25% by runtime
Enable Q8/Q4 KV quantization to scale batch concurrency
OLLAMA_CONTEXT_LENGTH=16384 ollama run qwen3:32bRTX 5090 Fits Qwen3-32B · Cost & TCO
Approx. $2,199 (~$69/GB VRAM) · 575W TDP
📊 Qwen3-32B Quantization VRAM Breakdown (16K Context)
| Quantization | Bits/Weight | Weights | KV Cache (16K) | Total VRAM | Quality Note |
|---|---|---|---|---|---|
| FP16 / BF16 | 16 bpw | 65.6 GB | 4.29 GB | 71.66 GB | Lossless Full Precision |
| FP8 / Q8_0 | 8.5 bpw | 34.85 GB | 4.29 GB | 40.29 GB | 99.9% Near-Lossless |
| GGUF Q6_K | 6.56 bpw | 26.9 GB | 4.29 GB | 32.18 GB | Ultra High Quality |
| GGUF Q5_K_M | 5.69 bpw | 23.33 GB | 4.29 GB | 28.54 GB | High Balanced Quality |
| Q4_K_M / AWQ | 4.85 bpw | 19.88 GB | 4.29 GB | 25.03 GB | Sweet Spot (Recommended) |
| GGUF IQ4_XS | 4.25 bpw | 17.42 GB | 4.29 GB | 22.52 GB | Compact 4-bit |
| GGUF Q3_K_M | 3.91 bpw | 16.03 GB | 4.29 GB | 21.1 GB | Moderate Perplexity Loss |
| GGUF Q2_K | 3.15 bpw | 12.91 GB | 4.29 GB | 17.92 GB | Extreme Compression |
🖥️ GPU & Apple Silicon Compatibility for Qwen3-32B (Q4_K_M)
Frequently Asked Questions
What are the exact architecture parameters of Qwen3-32B?
Qwen3-32B uses Qwen3ForCausalLM with 64 layers (64-Layer Dense Transformer (GQA + SwiGLU + QK-Norm)), hidden_size=5120, 64 Q heads / 8 KV heads (head_dim=128), and vocab_size=151,936.
How much VRAM does Qwen3-32B require?
At 4-bit quantization (Q4_K_M, 16K context), Qwen3-32B requires 25.03 GB VRAM. Full FP16 requires 71.66 GB.