Last updated:
Qwen
Alibaba's model family, shipped in an unusually wide range of sizes from mobile-friendly to a 2.4T-parameter flagship.
Model Facts
Latest model: Qwen3.8-27B
- Provider
- Alibaba
- Released
- Aug 14, 2026
- Context window
- 256K tokens (open models)
- Weights
- Open
Full Model List
| Model | Released | Status | Note |
|---|---|---|---|
| Qwen3.8-27B | Aug 14, 2026 | – | Apache 2.0 default local model |
| Qwen3.8-2.4T-A95B | Aug 12, 2026 | – | first open release of a Qwen-Max-class flagship, custom license |
| Qwen 3.7 | May 1, 2026 | – | closed-weight only (3.7-Max, 1M-token context) |
| Qwen 3.6 | Apr 1, 2026 | – | focused on agent workflows (27B and 35B-A3B open models) |
| Qwen 3.5 | Feb 1, 2026 | – | Apache 2.0 family with 256K native context |
| Qwen3-Coder | Jul 22, 2025 | legacy | agentic coding model |
| Qwen2.5-Coder | Nov 12, 2024 | legacy | code-specialized open models |
| Qwen2.5 | Sep 19, 2024 | – | widely used open-weight generation |
| Qwen2 | Jun 7, 2024 | – | broadened the size range and multilingual coverage |
| Qwen | Aug 3, 2023 | – | original release |
- Official model page:
- qwenlm.github.io
- Official documentation:
- qwen.readthedocs.io
- Other sources:
- Qwen 3.5 to 3.8 guide
What it is
Qwen is Alibaba's model family, available in an unusually wide range of sizes. The current generation is Qwen 3.8 (the 3.7 generation shipped closed-weight only). On August 12, 2026 Alibaba released open weights for Qwen3.8-2.4T-A95B, its first open Max-class flagship, and on August 14 it released Qwen3.8-27B under Apache 2.0 as the default local model.
Licensing
Qwen 3.5, 3.6 and Qwen3.8-27B are Apache 2.0. The 2.4T flagship uses a custom Qwen3.8-Max license with revenue-gated carve-outs, so check the terms before commercial use.
Where these numbers come from
The benchmarks are for Qwen3.8-27B: SWE-bench Pro 61.7, GPQA Diamond 89.2%, LiveCodeBench v6 90.3. The 256K context window is the native length across the 3.5 family; the closed Max models offer up to 1M tokens via API.
Notable traits
- Models from mobile-friendly to 2.4T parameters.
- A hybrid Gated DeltaNet architecture for long context.
- Mostly open weights, with a closed Max tier alongside.