Qwen3.8-Flash-Next (180B MoE) Architecture Parameters, Benchmarks & VRAM
Next-gen Qwen4Exp hybrid multimodal MoE (180B total / ~8B active, 512 experts).
🧬 Qwen3.8-Flash-Next (180B MoE) Architecture & Tensor Specifications
Verified directly against HuggingFace (Qwen/Qwen3.8-Flash-Next) config.json & Model Card
1. Backbone & Dimensions
qwen4_exp- Total / Active Params
- 180B / 8B
- Hidden Size (d_model)
- 2,560
- MoE Expert Dim
- 640
- MoE Experts
- 512 Routed · Top-10
- Activation / Norm
- SiLU (SwiGLU) · ε=0.000001
- Native Tensor Dtype
- BF16 / FP8
2. Attention & Layer Layout
HYBRID- Num Hidden Layers
- 48 Layers (12 Global KV)
- Hidden Layer Layout
- 48-Layer Sparse MoE (512 Experts, Top-10 Active)
- Attention Heads (Q / KV)
- 24 Q / 2 KV (12:1 GQA)
- Head Dimension
- 256
- Vision Encoder / MTP
- ViT 27L
3. Context, RoPE & Tokenizer
Tokenizer- Native Context Length
- 262,144 tokens
- RoPE Configuration
- RoPE (原生 262,144 Context)
- Vocab Size
- 248,320 (Untied)
📋 Qwen3.8-Flash-Next (180B MoE) Official API & System Card Specifications
Verified against official Alibaba Qwen developer documentation & pricing
1. Identity & Capabilities
API Spec- API Model ID
- qwen3.8-flash
- Release Date
- 2026-08-26
- 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
- 256,000 tokens (256K)
- Max Output Tokens
- 131,000 tokens
- Input Modalities
- Text + Image + Video
- Output Modalities
- Text
- Direct China Access
- ✅ Direct CN + RMB
- Deployment Mode
- Open-Weight + Cloud API
3. Official Pricing & Cache
CNY / 1M- Input / 1M Tokens
- ¥ 0.8
- Cached Input / 1M
- ¥ 0.1 -88%
- Output / 1M Tokens
- ¥ 2.7
- Tiered Pricing Notes
- Alibaba Bailian: Hosted on Alibaba Cloud Bailian (CN Direct)
🎛️ Official Sampling & Reasoning Parameters
Recommended generation settings & API parameter constraints
📊 Qwen3.8-Flash-Next (180B MoE) Full Benchmark Scorecard
22 public benchmarks · ranks among 154 models · synced 2026-11-20
💻 Coding 6 items
A multilingual benchmark for issue resolving in software engineering that covers Java, TypeScript, JavaScript, Go, Rust, C, and C++.
Vision2Web evaluates multimodal models on converting visual designs and screenshots into functional web pages, measuring end-to-end design-to-code capability.
SWE-Bench Pro is an advanced version of SWE-Bench that evaluates language models on complex, real-world software engineering tasks requiring extended reasoning and multi-step problem solving.
DeepSWE 1.1 evaluates software engineering agents using the mini-swe-agent harness.
RecreationBench is Qwen's long-horizon application-recreation benchmark for hybrid agents across Ubuntu, macOS, Windows, Android, and the web.
NL2Repo evaluates long-horizon coding capabilities including repository-level understanding, where models must generate or modify code across entire repositories from natural language specifications.
🤖 Agents & Tools 7 items
AndroidWorld evaluates an agent's ability to operate in real Android GUI environments, completing multi-step tasks by perceiving screen content and executing touch/type actions.
CoWorkBench is Qwen's internal cowork benchmark for evaluating long-horizon office and productivity agent tasks across domains such as computer science, finance, law, and medicine.
Tool Decathlon is a comprehensive benchmark for evaluating AI agents' ability to use multiple tools across diverse task categories. It measures proficiency in tool selection, sequencing, and execution across ten different tool-use scenarios.
ClawEval-MM is the multimodal variant of ClawEval, evaluating agentic problem solving with visual inputs.
Job Bench evaluates AI agents on realistic professional tasks that require multi-step planning, research, and production of work artifacts.
Agents' Last Exam is a challenging benchmark for AI agents on hard, long-horizon tasks that test sustained reasoning, planning, and tool use, reported with and without tool access.
OSWorld 2.0 is a benchmark of 108 long-horizon, real-world computer-use workflows spanning everyday and professional tasks.
📐 Math 2 items
MATH-Vision is a dataset designed to measure multimodal mathematical reasoning capabilities.
Humanity's Last Exam (HLE) is a multi-modal academic benchmark with 2,500 questions across mathematics, humanities, and natural sciences, designed to test LLM capabilities at the frontier of human knowledge with unambiguous, verifiable solutions.
👁️ Multimodal 4 items
CharXiv-R is the reasoning component of the CharXiv benchmark, focusing on complex reasoning questions that require synthesizing information across visual chart elements.
RealWorldQA is a benchmark designed to evaluate basic real-world spatial understanding capabilities of multimodal models.
LVBench is an extreme long video understanding benchmark designed to evaluate multimodal models on videos up to two hours in duration. It contains 6 major categories and 21 subcategories, with videos averaging five times longer than existing datasets.
Embodied Reasoning Question Answering benchmark consisting of 400 multiple-choice visual questions across spatial reasoning, trajectory reasoning, action reasoning, state estimation, and multi-view reasoning for evaluating AI capabilities in physical world interactions.
🧠 Reasoning & Science 2 items
LiveCodeBench is a holistic and contamination-free evaluation benchmark for large language models for code.
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.
📚 Knowledge & General 1 items
Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).
⚡ Qwen3.8-Flash-Next (180B MoE) Interactive VRAM & Token Speed Calculator
Physical 512 GB · Load 24.5%
MoE: reads only 4.85 GB active weights/step
Enable Q8/Q4 KV quantization to scale batch concurrency
OLLAMA_CONTEXT_LENGTH=16384 ollama run qwen3-8-flash-nextMac Studio M5 Ultra 512G Fits Qwen3.8-Flash-Next (180B MoE) · Cost & TCO
Approx. $9,499 (~$19/GB VRAM) · 190W TDP
📊 Qwen3.8-Flash-Next (180B MoE) Quantization VRAM Breakdown (16K Context)
| Quantization | Bits/Weight | Weights | KV Cache (16K) | Total VRAM | Quality Note |
|---|---|---|---|---|---|
| FP16 / BF16 | 16 bpw | 360 GB | 0.5 GB | 369 GB | Lossless Full Precision |
| FP8 / Q8_0 | 8.5 bpw | 191.25 GB | 0.5 GB | 196.88 GB | 99.9% Near-Lossless |
| GGUF Q6_K | 6.56 bpw | 147.6 GB | 0.5 GB | 152.35 GB | Ultra High Quality |
| GGUF Q5_K_M | 5.69 bpw | 128.03 GB | 0.5 GB | 132.39 GB | High Balanced Quality |
| Q4_K_M / AWQ | 4.85 bpw | 109.12 GB | 0.5 GB | 113.11 GB | Sweet Spot (Recommended) |
| GGUF IQ4_XS | 4.25 bpw | 95.63 GB | 0.5 GB | 99.34 GB | Compact 4-bit |
| GGUF Q3_K_M | 3.91 bpw | 87.98 GB | 0.5 GB | 91.54 GB | Moderate Perplexity Loss |
| GGUF Q2_K | 3.15 bpw | 70.88 GB | 0.5 GB | 74.09 GB | Extreme Compression |
🖥️ GPU & Apple Silicon Compatibility for Qwen3.8-Flash-Next (180B MoE) (Q4_K_M)
Frequently Asked Questions
What are the exact architecture parameters of Qwen3.8-Flash-Next (180B MoE)?
Qwen3.8-Flash-Next (180B MoE) uses Qwen4ExpForConditionalGeneration with 48 layers (48-Layer Sparse MoE (512 Experts, Top-10 Active)), hidden_size=2560, 24 Q heads / 2 KV heads (head_dim=256), and vocab_size=248,320.
How much VRAM does Qwen3.8-Flash-Next (180B MoE) require?
At 4-bit quantization (Q4_K_M, 16K context), Qwen3.8-Flash-Next (180B MoE) requires 113.11 GB VRAM. Full FP16 requires 369 GB.