Qwen3.6 35B A3B
Active Parameters
35B
Context Length
262K
Modality
Multimodal
Architecture
Mixture of Experts (MoE)
License
Apache 2.0
Release Date
15 Apr 2026
Knowledge Cutoff
-
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
4x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
262,144 tokens
Consumer
5x RTX 4090
24GB VRAM
Datacenter
2x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
Architecture Diagram
Evaluation Benchmarks
No evaluation benchmarks for Qwen3.6 35B A3B available.
Rankings
Overall Rank
-
Coding Rank
-
About Qwen3.6 35B A3B
Qwen3.6-35B-A3B is Alibaba's open-source hybrid MoE model with 35B total parameters and only 3B active per token. Built on a novel architecture combining Gated DeltaNet linear attention with standard Gated Attention and sparse MoE (256 experts, 8 routed + 1 shared active), it delivers exceptional agentic coding performance rivaling much larger dense models. Achieves 73.4% on SWE-bench Verified, 51.5% on Terminal-Bench 2.0, and 92.6% on AIME 2026. Natively multimodal (text, image, video), supports 262K context natively (up to 1M with YaRN), includes thinking preservation for agentic tasks, and is trained with Multi-Token Prediction. Available via Alibaba Cloud Model Studio API as qwen3.6-flash. Released April 15, 2026 under Apache 2.0.
Technical Specifications
Attention
Attention Structure
Grouped-Query Attention
Attention Heads
16
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
Hidden Dimension Size
2,048
Number of Layers
40
FFN Intermediate Size (Dense)
512
Multi-Token Prediction Heads
1
Tokenizer
Vocabulary Size
248,320
Mixture of Experts
Total Expert Parameters
3.0B
Number of Experts
256
Active Experts
9
Shared Experts
-
FFN Intermediate Size (per Expert)
512
Dense Layers Before MoE
-
About Qwen 3.6
Qwen 3.6 is Alibaba's latest generation of hybrid sparse Mixture-of-Experts (MoE) models featuring a novel architecture that combines Gated DeltaNet linear attention layers with standard Gated Attention layers and MoE feed-forward networks. The family delivers substantial improvements in agentic coding, multimodal perception, and reasoning, with native support for thinking and non-thinking modes, thinking preservation across turns, and a 262K native context window.
Other Qwen 3.6 Models
- No related models available