Qwen3.6
Alibaba's Apache 2.0 Qwen3.6 pair: a 35B mixture of experts with 3B active and a 27B dense model, tuned for agentic coding.
In short. Qwen3.6 is a pair of free AI models from Alibaba that are especially good at working as coding assistants. One is quick because it only uses 3 billion of its 35 billion parameters at a time; the other is a single 27B model that fits one good graphics card once compressed. Both read images and video, and the Apache 2.0 license allows commercial use.
Summary
Alibaba's Qwen team released Qwen3.6-35B-A3B on 16 April 2026 and Qwen3.6-27B on 22 April 2026 under the Apache 2.0 license, following the Qwen3.5 series from February. Qwen says it built Qwen3.6 on direct community feedback, with the priorities of stability and real-world usefulness. The upgrades it highlights are agentic coding, meaning front-end workflows and reasoning across a whole code repository, and a new thinking-preservation option that keeps the reasoning from earlier messages in the conversation history, so iterative work does not repeat itself.
The two models share the Qwen3.5 design: layers alternate Gated DeltaNet, a linear-attention design, with gated attention in a 3-to-1 pattern, and both accept text, images and video while writing text. The 35B-A3B is a mixture-of-experts model with 40 layers and 256 experts, using 8 routed experts plus 1 shared expert per token, so about 3 billion of its roughly 36 billion parameters are active. The 27B is a dense model with 64 layers. Both have a 262,144-token native context that the model cards say can be extended to about 1,010,000 tokens, and both were trained with multi-token prediction.
BF16 downloads are about 71.9 GB for the 35B-A3B and 55.6 GB for the 27B; the official FP8 versions are about 37.5 GB and 30.9 GB. The weights work with Transformers, vLLM, SGLang and KTransformers. Qwen's own table puts the 27B at 77.2 on SWE-bench Verified and 59.3 on Terminal-Bench 2.0, against 75.0 and 41.6 for Qwen3.5-27B, and the 35B-A3B at 73.4 and 51.5; these results are self-reported and use Qwen's own agent setups.
Good for
- Local coding assistants and repository-level agents on a single 24 to 48 GB GPU or a high-memory Mac, using the 27B
- Fast token generation on modest hardware with the 35B-A3B, as long as the full weights fit in memory
- Long agent sessions that benefit from keeping earlier reasoning in the history
- Commercial products, since Apache 2.0 has no revenue limit
Watch out for
- Qwen3.8-27B is newer than Qwen3.6-27B at the same size, so compare the two on your own tasks
- The 35B-A3B is quick because only 3B parameters are active per token, but all 35.95B must still sit in memory
- Long contexts need much more memory than the weights alone
- The benchmark numbers are Qwen's own and depend on its agent setups
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Estimates for the 27B variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.
| Variants | Parameters | BF16 | FP8 / INT8 | 4-bit (Q4) | Fits on |
|---|---|---|---|---|---|
| 27B BF16 | 27.8B | 64 GB | 32 GB | 19 GB | Gaming GPU |
| 27B FP8 FP8 | 27.8B | 64 GB | 32 GB | 19 GB | Gaming GPU |
| 35B-A3B BF16 | 36B 3B active | 83 GB | 41 GB | 25 GB | Gaming GPU |
| 35B-A3B FP8 FP8 | 36B 3B active | 83 GB | 41 GB | 25 GB | Gaming GPU |
Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.
The license in plain words
One of the most widely used permissive licenses. You may use, change and sell products built on the model with only light conditions.
See also
- Nemotron 3.5 Lightning vs Qwen3.6: which 3B-active model to use?
- Ornith 1.5 vs Qwen3.6: which agentic coding model should you use?
- Best open-source models for coding and AI agents
- Best open-source multimodal models in 2026 (images, video and audio)
- Best small open models you can run on a laptop
- Raon-1: KRAFTON's Korean-first 28.8B model and a 3.1B on-device sibling
Frequently asked questions
- What is Qwen3.6?
- Qwen3.6 is a pair of free AI models from Alibaba that are especially good at working as coding assistants. One is quick because it only uses 3 billion of its 35 billion parameters at a time; the other is a single 27B model that fits one good graphics card once compressed. Both read images and video, and the Apache 2.0 license allows commercial use.
- Can I use Qwen3.6 commercially?
- Yes. Qwen3.6 is released under Apache 2.0, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
- Is Qwen3.6 open source?
- Qwen3.6 is released under Apache 2.0, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
- Can I run Qwen3.6 on my own computer?
- Yes, on a good graphics card. The smallest variant, 27B, needs about 19 GB of memory at 4-bit precision, which suits a 24 to 32 GB gaming GPU or a Mac with 32 GB or more.
- When was Qwen3.6 released?
- Qwen3.6 was released by Alibaba Qwen on April 16, 2026.
- How long a context window does Qwen3.6 have?
- Qwen3.6 supports a context window of up to 256K tokens.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- Qwen repository and release news (GitHub) github.com
- Qwen3.6-35B-A3B release blog (Qwen) qwen.ai
- Qwen3.6-27B release blog (Qwen) qwen.ai
- Qwen3.6-35B-A3B model card (Hugging Face) huggingface.co
- Qwen3.6-27B model card (Hugging Face) huggingface.co
Release date and facts last verified November 11, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| Qwen3.5 Alibaba Qwen | 873M – 403B 17B active | 256K | Apache 2.0 (Permissive) | 0.602 GB Laptop or phone | |
| Qwen3.8-27B Alibaba Qwen | 27.8B | 256K | Apache 2.0 (Permissive) | 19 GB Gaming GPU | |
| Gemma 4 Google DeepMind | 5.1B – 30.7B | 256K | Apache 2.0 (Permissive) | 3.52 GB Laptop or phone | |
| GLM-5.3-Flash Z.ai (Zhipu AI) | 320B 18B active | 1M | MIT (Permissive) | 221 GB Big-memory desktop |