Qwen3.6-27B

Qwen Team (Alibaba Cloud)

Parameters

27.0B

Architecture

Hybrid Gated DeltaNet + Gated Attention (dense)

Released

21.04.2026

License

Apache License 2.0

Open Weights Commercial Use Multimodal BF16 Qwen3.6 en zh multi

Input Modalities

text image video

Output Modalities

text

Context (native)

262,144 tokens

Context (extended)

1,010,000 tokens

Openness Index Score 100.0/100

About

Qwen3.6-27B (Qwen/Qwen3.6-27B) is the dense mid-size multimodal model of Alibaba's Qwen3.6 line - 27B parameters (dense; FFN-based, no MoE) handling text, image and video input. Its 64-layer hybrid stack repeats 16x (3x (Gated DeltaNet -> FFN) + 1x (Gated Attention -> FFN)): 48 linear-attention layers (Gated DeltaNet, 48 V / 16 QK heads, head dim 128) and 16 full-attention layers (Gated Attention, 24 Q / 4 KV heads, head dim 256, RoPE dim 64), with hidden size 5120, FFN intermediate dimension 17408, 248K vocabulary, Multi-Token Prediction (MTP) trained with multi-steps, and a 262,144-token native context extensible to 1,010,000.

The Qwen3.6 release delivers substantial upgrades in agentic coding - frontend workflows and repository-level reasoning with greater fluency and precision - and introduces thinking preservation, a new option to retain reasoning context from historical messages that streamlines iterative development and reduces overhead. Released April 21, 2026 under Apache 2.0.

Training Data Pre-training & Post-training. Trained with multi-token prediction (MTP) steps. Post-trained with Thinking Preservation feature.

Benchmark Scores

Benchmark Score Date Standing
SWE-bench Verified
coding_agent
88.07%
22.04.2026
#16/60
better than 75%
SWE-bench Multilingual
coding_agent
78.46%
22.04.2026
#25/48
better than 49%
Terminal-Bench 2.0
coding_agent
74.86%
22.04.2026
#8/23
better than 64%
SkillsBench Avg5
coding_agent
70.30%
22.04.2026
#9/16
better than 47%
QwenWebBench
coding_agent
91.22%
22.04.2026
#2/8
better than 86%
NL2Repo
coding_agent
39.85%
22.04.2026
#27/41
better than 35%
Claw-Eval Avg
coding_agent
87.45%
22.04.2026
#7/40
better than 85%
QwenClawBench
coding_agent
100.00%
22.04.2026
#1/17
better than 100%
MMLU-Pro
knowledge
84.52%
22.04.2026
#12/48
better than 74%
MMLU-Redux
knowledge
91.18%
22.04.2026
#8/30
better than 76%

Showing 10 of 78 results

What is a good SWE-bench Verified score? What is a good SWE-bench Multilingual score? What is a good Terminal-Bench 2.0 score?

Architecture

Decoder Block ×64 input Embedding vocab 248K · d 5120 Linear / Recurrent Hybrid 24:4 · dₕ 256 ×48 Full Attention Hybrid 24:4 · dₕ 256 ×16 Dense FFN silu · d 17K Final Norm LM Head vocab 248K output
Attention
Hybrid Attention (24:4)
Layers
64
Hidden size
5120
Context
262K tokens
Parameters
27000M

Source: Hugging Face config.json · Qwen3_5ForConditionalGeneration · exact layer pattern · model repo

Type: Hybrid Dense Transformer
Attention: 16x(3x(Gated DeltaNet) + 1x(Gated Attention)); Gated DeltaNet 48 V / 16 QK heads, Gated Attention 24 Q / 4 KV heads
Decoder: Causal Language Model with Vision Encoder
Token Embedding

248K

Training Pipeline

  1. 1
    pretraining

    Pre-training

    Pre-training of the dense 27B language model with hybrid Gated DeltaNet + Gated Attention architecture and multi-token prediction (MTP).

  2. 2
    other

    Post-training

    Post-training including thinking-mode alignment and introduction of Thinking Preservation (retaining reasoning traces across multi-turn conversations).

Training & Evaluation Datasets

NameRoleSizeModalitiesCollection
Qwen3.6 pre-training corpus (multimodal) pretraining — —

Linked Resources

website

Qwen3.6-27B Model Card (Hugging Face)

https://huggingface.co/Qwen/Qwen3.6-27B

 blog

Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model

https://qwen.ai/blog?id=qwen3.6-27b

 github

QwenLM/Qwen-Agent

https://github.com/QwenLM/Qwen-Agent

 github

QwenLM/qwen-code

https://github.com/QwenLM/qwen-code

 demo

Qwen Chat

https://chat.qwen.ai/

 collection

Qwen3.6 Collection

https://huggingface.co/collections/Qwen/qwen36

Comments (0)

No comments yet

Be the first to share your thoughts!

What now?

You have read the full card for Qwen3.6-27B — here is where to go next.

Compare with another model Test on your own GPU Related models Data as JSON