Strengths
- It scored 58 on the Artificial Analysis Intelligence Index in August 2026, fifth among all models.
- Strong with images: second in LMArena Vision Arena with 1,301 Elo, right behind Claude Fable 5.
- A one-million-token context and text, image and video input for $2 input and $6 output per million tokens.
- Best result among tested models on tau3-Banking, a test of agents doing banking tasks (51.3%).
- You switch thinking on only when you need it, and the price is the same either way.
Weaknesses
- It writes more tokens than rivals, so one task in the Artificial Analysis test cost $1.13, more than the low per-token price suggests.
- Alibaba has not published a technical report or model card. Its knowledge cutoff and training data are unknown.
- In Frankfurt it only runs in Global scope, so processing is not tied to the EU.
- You cannot download the API model with vision and long context, only its text-only base is public.
Best for
- agentic coding
- analysing images and video
- long documents up to one million tokens
- a capable API model at a lower price
Not ideal for
- data that may only be processed in the EU
- companies that need detailed model documentation
How to use it in Europe
Through the API in Alibaba Cloud Model Studio. The international Singapore region gives you a free quota of 1 million tokens for 90 days. In Frankfurt your data is stored in Germany, but Max only runs there in Global scope, so processing may happen outside the EU. For consumers the model is in the QwenWork app with plans in USD, and Alibaba does not say whether you can subscribe from Slovakia or Czechia. If data must stay in house, use the open Qwen3.8-2.4T-A95B on your own server.
What is coming
Confirmed by vendor At its Apsara conference in September 2026, Alibaba confirmed it is training Qwen 4. It has not announced a release date, prices or a licence. When: no date yet, Alibaba has only said the model is in training →
Price compared with similar models
Benchmark results
| Benchmark | Score | Source |
|---|---|---|
| Artificial Analysis Intelligence Index | 58 (#5 overall) | deeplearning.ai ↗, August 2026 |
| LMArena Vision Arena | 1301 Elo (#2) | deeplearning.ai ↗, August 2026 |
| Terminal-Bench 2.1 | 86.6 | huggingface.co ↗ (vendor-reported), August 2026 |
| SWE-bench Pro | 67.7 | huggingface.co ↗ (vendor-reported), August 2026 |
| GPQA Diamond | 92.6 | 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
- Qwen3.7-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.
- Qwen3.8-27B: A small open Qwen model under Apache 2.0 that runs on your own computer and understands images and video.
Similar models from other vendors
- Claude Opus 5.5 (Anthropic): Anthropic's main model for demanding work, coding and long documents, and the one Anthropic recommends starting with.
- DeepSeek V4 Pro (DeepSeek): DeepSeek's largest open-weight model for demanding text tasks at a low price. It does not understand images.
- Gemini 4 Carbon (Google): Google's most capable model, announced on 28 October 2026. So far only selected cybersecurity experts have access.
- Gemini 4 Argon (Google): Google's most capable publicly available model, out in October 2026. Built for demanding analytical work where accuracy matters and made-up answers are costly.
- Muse Wonder (Meta): Meta's strongest model for long-running agents and coding, still released as a preview.
- Mistral Large 4 (Mistral AI): Mistral's largest model, still in preview. Built for demanding tasks, long documents and security analysis with data kept in the EU.