Active Parameters
176B
Context Length
1M
Modality
Multimodal
Auxiliary Parameters
450M
Architecture
Mixture of Experts (MoE)
License
other
Release Date
24 Aug 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Input: $0.15 · Output: $0.47
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
18x RTX 4090
24GB VRAM
Datacenter
5x NVIDIA A100
80GB VRAM
Apple Silicon
4x Apple M3 Max
128GB VRAM
1,000,000 tokens
Consumer
23x RTX 4090
24GB VRAM
Datacenter
7x NVIDIA A100
80GB VRAM
Apple Silicon
5x Apple M3 Max
128GB VRAM
Rank
#38
| Benchmark | Score | Rank |
|---|---|---|
Agentic Index | 0.54 | 5 |
Agentic Coding | 0.62 | 10 |
Web Development | 1636 | 11 |
Graduate-Level QA | 0.917 | 18 |
Reasoning | 0.87 | 23 |
General | 0.76 | 25 |
LiveBench Average | 0.76 | 25 |
Coding Index | 0.73 | 30 |
Data Analysis | 0.74 | 31 |
Intelligence Index | 0.40 | 40 |
Mathematics | 0.86 | 43 |
Coding | 0.73 | 44 |
Overall Rank
#38
Coding Rank
#44
Qwen 3.8 Flash is an efficient multimodal reasoning model from Alibaba designed for rapid inference across coding, agentic workflows, and document analysis. It delivers strong vision-language understanding while maintaining low latency and a long context window.
Attention
Attention Structure
DeepSeek Sparse Attention
Attention Heads
24
Key-Value Heads
2
Attention Head Dimension
256
Position Embedding
ROPE
RoPE Theta
10,000,000
Sliding Window Attention
No
Sliding Window Size
-
Sliding Window Ratio
-
Linear Attention
Yes
Linear Attention Ratio
75.0%
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Auxiliary Parameters
450M
Hidden Dimension Size
2,560
Number of Layers
48
FFN Intermediate Size (Dense)
-
Multi-Token Prediction Heads
1
Tokenizer
Vocabulary Size
248,320
Mixture of Experts
Total Expert Parameters
6.0B
Number of Experts
512
Active Experts
10
Shared Experts
1
FFN Intermediate Size (per Expert)
640
Dense Layers Before MoE
0
Alibaba's Qwen 3.8 generation represents the frontier hybrid Mixture-of-Experts architecture designed for coding, professional work, research, and long-horizon agentic tasks. It features a 2.4-trillion parameter architecture (95B active per token) combining Gated DeltaNet linear attention with standard Gated Attention, available both as open weights and as a hosted flagship service.
Assistant
Online