Qwen 3.8 2.4T A95B
Active Parameters
2.4T
Context Length
1.05M
Modality
Text
Architecture
Mixture of Experts
KV Cache Rate
368 KB/tok
License
other
Release Date
8 Aug 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Input: $2.00 · Output: $6.00
Hardware Sizing
Precision
Context
Workload Memory: 4652.67 GB
FP16 · 8K context
Base Weights: 4647.30 GB
KV Cache: 2.88 GB
Runtime & Graphs: 2.50 GB
Minimum Required Hardware
Workload exceeds 4,000 GB memory ceiling. Custom multi-node cluster orchestration required.
Open in VRAM Calculator
Determine requirements for your specific workload:
Open in VRAM Calculator
Architecture Diagram
Evaluation Benchmarks
Rank
#16
| Benchmark | Score | Rank |
|---|---|---|
Agentic Index | 0.50 | 19 |
Intelligence Index | 0.40 | 29 |
Coding Index | 0.72 | 34 |
Rankings
Overall Rank
#16
Coding Rank
#32
About Qwen 3.8 2.4T A95B
Qwen 3.8 2.4T A95B is a massive open-weight sparse mixture-of-experts model from Alibaba activating 95B parameters out of 2.4T total. It provides frontier-level general reasoning, coding, and multilingual performance as the open variant of Qwen 3.8 Max.
Technical Specifications
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
64
Key-Value Heads
4
Attention Head Dimension
256
Position Embedding
ROPE
RoPE Theta
10,000,000
Linear Attention
Yes
Linear Attention Ratio
75.0%
Normalization
RMS Normalization
Activation Function
SwigLU
Dimensions
Hidden Dimension Size
8,192
Number of Layers
92
Multi-Token Prediction Heads
1
Tokenizer
Vocabulary Size
248,320
Mixture of Experts
Total Expert Parameters
95.0B
Number of Experts
512
Active Experts
10
Shared Experts
1
FFN Intermediate Size (per Expert)
2,048
Dense Layers Before MoE
0
Resources
About Qwen 3.8
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.