Alibaba
Updated May 2026
Qwen
Alibaba's open-weights family that leads the Chinese lineup
Qwen is the output of Alibaba Cloud's research team, and it ships faster than almost any frontier lab — regularly publishing new open-weight models every month. As of April 2026 their flagship is Qwen 3.6 Plus with 1M context and agentic coding abilities, and their VL (vision-language) and Coder variants lead open benchmarks. Qwen's license is genuinely commercial-friendly, making it Llama's biggest open-weight rival.
Variants
6
Best at
extremely fast release cadence
Max context
1M
tokens
Pricing
Free tier
$0
per Alibaba Cloud Model Studio trial
Qwen 3.6 Plus API
~$0.80 in / $3.20 out
per million tokens
Self-host
$0
per weights on HuggingFace/ModelScope
Variants
Sort the table by context window or cost to find the right variant. Click any version below for a battle card with ranks, pricing notes, and official links.
| Modalities | ||||
|---|---|---|---|---|
Qwen 3.6 Plus qwen-3-6-plus | 1M | varies / varies | 2026 | textvisioncode |
Qwen3-Coder qwen3-coder | 1M | varies / varies | 2025 | code |
Qwen 3.6 35B A3B qwen-3-6-35b-a3b | 262K | varies / varies | 2026 | text |
Qwen 3.5 122B A10B qwen-3-5-122b-a10b | 262K | varies / varies | 2026 | text |
Qwen3-VL 235B qwen3-vl-235b | 128K | varies / varies | 2025 | textvision |
Qwen 2.5 qwen-2-5 | 128K | varies / varies | 2024 | textvision |
Battle card
Context rank
#1
within Qwen
Capability rank
#1
modalities + reasoning
Weights
Open
self-hostable if licensed
Best fights to pick
- autonomous coding agents
- whole-codebase refactors
- computer-use automation
Rankings are Tendril directory ranks, computed from the model data shown here. Public benchmark leaderboards change often, so official docs and current benchmark pages should be checked before buying or deploying.
Learn
Lessons about this model
Structured lessons that cover Qwen directly or put it in context alongside its rivals.
Check yourself
Quizzes
Short, mixed-difficulty quiz sets on Qwen and its model family.
Open-Weight Families: Llama, Mistral, Qwen, DeepSeek, Gemma
7 questions
The open ecosystem that shook the industry.
Start quiz →Hands-on
Try these prompts
Ready-made prompts that show Qwen at its best. Use them in your own AI workspace, then compare the output with what you learned in Tendril.
Chinese-English code comments
BuildersQwen's bilingual training makes it unusually good at code with mixed-language documentation.
Translate the docstrings in this Python file from Chinese to English while keeping the code untouched. Preserve any technical jargon that doesn't translate cleanly and flag it.
Qwen VL describing a photo
CreatorsHand Qwen VL an image and ask for a structured description.
[Attach photo] Describe this image for a screen-reader user. Include: main subject, setting, mood, any text visible, and anything a sighted person might miss on first glance.
Qwen Coder on a small feature
BuildersA focused coding ask — good for testing Qwen's ability to ship production-shaped snippets.
Write a Next.js API route that accepts a POST with { email, message } and forwards it to a Resend contact form. Include input validation, rate limiting via a simple in-memory counter, and error handling.Print & keep
Printable reference
One-page summaries and flowcharts — great for desks, classrooms, or study sessions.
Go deeper
Official resources
Straight from the lab — docs, API references, and the chat surfaces you can try today.
Strengths
- extremely fast release cadence
- open weights across small-to-large sizes
- strong vision and code variants
Limits
- PRC censorship applies
- English polish trails Claude/GPT
- Alibaba Cloud service has data-residency concerns abroad