Qwen (Alibaba) — the model line at a glance
Where things stand
As of September 2026, Qwen 3.8-Max (August 2026) is the flagship of the line — and the first Max model with open weights. This article covers the line as a whole; figures for individual versions live in the linked glossary entries.
What Qwen is and what the line stands for
Qwen is Alibaba’s model line — and the only one of the big Chinese lines that runs two tracks at once. On one side an open dense/MoE series under Apache 2.0 that you can freely download and self-host. On the other a proprietary Max line that has so far only been accessible via API. This dual structure is Qwen’s defining feature: one of the broadest ranges of size tiers anywhere, combined with a closed top model.
The current flagship Qwen 3.8-Max blurs that line for the first time: a MoE model with 2.4 trillion parameters and a 1-million-token context — and, as the first Max model, also with open weights. Its predecessors Qwen3-Max and Qwen 3.7-Max stayed pure API models; the open Qwen line ran separately during that time and only picked up the open-weights tradition at Max level again with 3.8-Max.
The line’s strength profile
- Coding 4 / 5 · Claude Opus 5.5 u. a.
- Reasoning 4 / 5 · Claude Opus 5.5 u. a.
- Text 4 / 5 · Claude Opus 5.5 u. a.
- Vision 4 / 5 · Gemini 3.1 Pro
- Speed 3.5 / 5 · Claude Haiku 4.5 u. a.
- Kosten-Eff. 4 / 5 · GPT Luna u. a.
Eignung 0–5 · redaktionelle Einordnung, kein Benchmark · gestrichelt = Feld-Bestwert je Achse
Qwen is a balanced all-rounder across the axes, with no single peak — but with the widest range of model sizes underneath. The axes are an editorial assessment, not a benchmark — they show the balance, but don’t replace a test on your own use case.
Where Qwen sits in the field
Redaktionelle Einordnung, kein Benchmark
Next to Kimi, GLM and DeepSeek V, Qwen sits in the solid all-rounder band with a good cost profile. What sets Qwen apart doesn’t show on the map: the widest range of size tiers and the tight integration into Alibaba Cloud.
Use profile — from edge to frontier, open and proprietary
Qwen is at its best where variety and multilingual coverage matter:
- Self-hosting across many size tiers — from small models for edge devices to the large MoE flagship. The open series sits under Apache 2.0 on Hugging Face.
- Multilingual applications with a Mandarin focus, where Qwen is the strongest open option.
- Cloud-native projects in the Alibaba ecosystem.
- Agentic workflows with long context, thanks to the 1M-token window on the flagship.
For workloads that need full control over your own infrastructure, the open dense/MoE series is the right address; the Max line covers the case where API access to the top model is enough. Self-hosting basics are under Running local LLMs.
Which Qwen version fits
For new projects the current flagship Qwen 3.8-Max is the right pick — open and powerful at once. If you need the API route or older size tiers, the right version is here:
- Qwen 3.8-Max — current flagship, open (August 2026).
- Qwen 3.7-Max and Qwen 3-Max — pure API models, without open weights.
- Qwen 3.6, Qwen 3.5, Qwen 3 and Qwen 2.5 — open series, earlier generations.
Related terms
- AI model families at a glance — where Qwen sits next to the other open lines.
- Running local LLMs — self-hosting across many size tiers.
- The Hugging Face ecosystem — where the open weights live.
FAQ
- Both. Alibaba runs two tracks: an open dense/MoE series under Apache 2.0 and a proprietary Max line accessible only via API. Qwen 3.8-Max is so far the only exception with both access routes.
- Qwen 3.8-Max, released in August 2026 — a MoE model with 2.4 trillion parameters, a 1M-token context and, as the first Max model, also open weights.
- No. Both are pure API models without published weights. The open Qwen line ran separately during that time and only picked up the open-weights tradition at Max level again with Qwen 3.8-Max.
- Self-hosted deployments across many size tiers, multilingual applications with a Mandarin focus, cloud-native projects in the Alibaba ecosystem and agentic workflows with long context.
- All four are open Chinese frontier lines. Qwen stands out through the widest range of size tiers and Alibaba Cloud integration, while Kimi leads on context window and agent swarm, GLM on coding benchmarks and DeepSeek on price-performance.