Strengths
- The permissive Apache 2.0 license allows commercial use, and the model is small enough for a powerful consumer machine.
- It scored 52 on the Artificial Analysis Intelligence Index in August 2026, third among open-weight models.
- It accepts images and video, and thinking can be switched off for each request.
- Alibaba reports 61.7 on the SWE-bench Pro coding test, only 6 points below Qwen3.8-Max.
Weaknesses
- According to DeepLearning.AI, it sometimes overthinks.
- Thinking is on by default at the high xhigh effort level, which makes waits and answers longer until you switch it off.
- It natively handles 262,144 tokens. Reaching one million requires extending it with YaRN, which takes some setup.
Best for
- a local assistant without internet
- deployment on company servers
- fine-tuning for your own tasks
- sensitive data that must not go to the cloud
Not ideal for
- tasks that need the broad knowledge of the largest models
How to use it in Europe
You can download the weights for free from Hugging Face and use them commercially under Apache 2.0. The model runs fully offline on your own computer or server, so your data goes nowhere. Alibaba does not publish exact hardware requirements. If you would rather not install anything, a hosted version is available in Alibaba Cloud Model Studio in the Singapore region.
Price compared with similar models
Benchmark results
| Benchmark | Score | Source |
|---|---|---|
| Artificial Analysis Intelligence Index | 52 (#3 among open-weights models) | deeplearning.ai ↗, August 2026 |
| SWE-bench Pro | 61.7 | huggingface.co ↗ (vendor-reported), August 2026 |
| LiveCodeBench v6 | 90.3 | huggingface.co ↗ (vendor-reported), August 2026 |
| GPQA Diamond | 89.2 | huggingface.co ↗ (vendor-reported), August 2026 |
Benchmarks cover only a narrow slice of ability and vendors often report them at their best settings. Treat them as a rough guide.
More models from this vendor
- Qwen3.8-Max: Alibaba's flagship with a one-million-token context that understands text, images and video. Strong at coding and fairly priced.
- Qwen3.7-Plus: An affordable mid-range Qwen model for everyday business tasks through the API, with a one-million-token context and image input.
- Qwen3.8-Flash: A very cheap, fast Qwen model for high volumes of tasks. Unlike Max and Plus, it runs in Frankfurt with processing kept in the EU.
- Qwen3.8-2.4T-A95B: The open-weight text base of Qwen3.8-Max with 2.4 trillion parameters, for large companies with their own GPU cluster.
Similar models from other vendors
- Gemma 4 31B (Google): Google's largest open model, which you can run on your own computer or server when data must not leave the company.
- Gemma 4 26B A4B (Google): Google's faster open model for your own server. Close to the larger Gemma 4 31B in quality, but quicker to respond.
- Muse Glimmer 30B (Meta): Meta's small open model under Apache 2.0 that runs on a single graphics card, even offline.
- Llama 4 Scout (Meta): The smaller Llama 4 from April 2025, very cheap through providers. Its license grants no rights to people or companies in the EU.
- Ministral 3 14B (Mistral AI): Mistral's small open model under Apache 2.0 that runs on a single GPU, even without an internet connection.
- Nemotron 3.5 Bolt 74B-A9B (NVIDIA): NVIDIA's open model for AI agents on your own powerful PC, small enough to fit in 24 GB of graphics card memory.