Alibaba Cloud Open-Weight Models
The most-forked open-weight model family in AI. 90,000+ derivatives. Apache 2.0. From $0.15/M tokens to frontier-tier performance at $2.50/M — Qwen spans every budget.
Qwen3.7-Max
~1T params, 1M context, $2.50/M input
Qwen3.6-35B-A3B
Apache 2.0, $0.15/M input, open-weight
Qwen Code
Terminal agent, 35-hr autonomous sessions
201 Languages
Broadest multilingual support in open AI
Qwen3-235B-A22B
Top-5 globally: SWE-Bench Pro 60.6% & Terminal-Bench 2.0 69.7%
May 2026
Qwen3.7-Max Launch
~1 trillion parameters with Hybrid Gated DeltaNet. 1M token context window. 3:1 linear-to-full attention ratio for efficient long-context inference.
ReleaseApr 2026
SWE-Bench Pro 60.6%
Qwen3-235B-A22B scores 60.6% on SWE-Bench Pro and 69.7% on Terminal-Bench 2.0. Top-5 globally on both coding leaderboards.
BenchmarkMar 2026
Qwen Code: 35-Hour Session
Verified 35-hour autonomous coding run with 1,158 tool calls. Native MCP server support ships alongside the terminal agent.
MilestoneQ1 2026
90,000+ Derivative Models
ModelScope records 90,000+ models derived from Qwen weights. Apache 2.0 licensing on sub-35B models powers the ecosystem.
Open WeightAlibaba Cloud's Qwen (Tongyi Qianwen division) launched in beta April 2023 and hit public release September 2023. Today it is the most-forked open-weight family on ModelScope, spanning 201 languages and two licensing tiers.
Models at 35B parameters and below ship under Apache 2.0 — unrestricted commercial fine-tuning and self-hosting, no royalties. The 72B+ tier uses the Tongyi Qianwen License, allowing self-hosting but restricting redistribution. Over 90,000 derivative models on ModelScope confirm the ecosystem is producing real-world results, not just download counts.
Qwen3.6-35B-A3B starts at $0.15/M input tokens, with output at $1.00/M. Frontier-tier Qwen3.7-Max runs $2.50/M input and $7.50/M output, with cached inputs dropping 90% to $0.25/M. The pricing ladder means a development budget can use 35B for iteration and Max for production without switching providers.
Qwen3.7-Max uses Hybrid Gated DeltaNet — a 3:1 ratio of linear to full attention layers. This lets the model process 1M token contexts without the quadratic memory cost of pure attention transformers. The architecture makes million-token tasks practical at the API tier, not just for self-hosted research clusters.
~1T
Max Params (Qwen3.7)
201
Languages Supported
60.6%
SWE-Bench Pro Score
$0.15
35B Input / MTok
May 2026
Qwen3.7-Max
~1 trillion parameters. Hybrid Gated DeltaNet with 3:1 linear-to-full attention. 1M token context. Priced at $2.50/M input, $7.50/M output, with 90% cached input discount.
Early 2026
Qwen3-235B-A22B & Qwen Code
Frontier MoE model scores top-5 globally: SWE-Bench Pro 60.6%, Terminal-Bench 2.0 69.7%. Qwen Code ships with native MCP server support and logs a 35-hour, 1,158-tool-call autonomous coding session.
Q4 2025
Qwen3.6-35B-A3B — Apache 2.0 Tier
Open-weight release priced at $0.15/M input, $1.00/M output. Apache 2.0 license enables unrestricted commercial fine-tuning for sub-35B models. Derivative model count crosses 90,000 on ModelScope.
2024
Qwen2 — 201 Language Expansion
Major multilingual expansion to 201 supported languages. Tongyi Qianwen License introduced for 72B+ parameter tier. Ecosystem growth accelerates on ModelScope.
Sep 2023
Qwen Public Launch
Alibaba Cloud's Tongyi Qianwen reaches public availability. First open-weight releases establish the derivative model ecosystem that grows to 90,000+.
Apr 2023
Tongyi Qianwen Beta
Alibaba Cloud's Tongyi Qianwen division launches closed beta. Foundation for what becomes the world's most-forked open-weight model family.
In-depth coverage of Qwen's architecture, API pricing, local deployment, and head-to-head comparisons. Open-weight AI analyzed with verified benchmarks and honest trade-offs.
Alibaba Cloud's open-weight juggernaut explained: architecture, capabilities, and why 90,000+ models have been built on it.
Free tiers, API pricing, cached discounts up to 90%, and enterprise deployment options compared.
The two Chinese labs competing for open-weight dominance. Benchmark showdown, pricing, and which one fits your stack.
Ollama, llama.cpp, and vLLM setup — hardware requirements, thinking mode, OpenAI API integration, and IDE connections for every Qwen3 model size.
Five integration paths — DashScope direct, OpenRouter, OpenAI SDK, thinking mode toggle, and MCP — with working Python code examples for each.
Compare Qwen against its closest open-weight and frontier-tier rivals, or explore the broader AI Tools Hub.
DeepSeek Hub
China's reasoning-focused rival with MATH-500 dominance.
Mistral Hub
European open-weight alternative with Apache 2.0 pedigree.
Meta Llama Hub
The open-weight OG — largest derivative ecosystem before Qwen.
Anthropic Claude Hub
The safety-first frontier model Qwen benchmarks against.
AI Tools Hub
65+ articles across 11 vendors. Breakdowns, comparisons, and guides.
AI Governance
Responsible AI, EU AI Act, and compliance frameworks.
Important context for responsible AI adoption
Qwen API is operated by Alibaba Cloud. Data processed through hosted endpoints is subject to Alibaba Cloud's terms of service and applicable Chinese data protection regulations. The free-tier interface and standard API may log conversations for model improvement. Enterprise customers on dedicated instances and developers running self-hosted open-weight models under Apache 2.0 have direct control over data residency. Review Alibaba Cloud's current privacy policy before processing confidential or personally identifiable information through any hosted endpoint.
AI assistants can create patterns of over-reliance. Qwen models are designed for information retrieval, coding, and multilingual tasks — not as substitutes for human expertise or emotional support. If you are experiencing distress:
AI systems can produce plausible-sounding but incorrect guidance. For mental health, medical, legal, or financial decisions, always consult a qualified professional.
See the NIST AI Risk Management Framework for structured guidance on AI risk assessment.
Under GDPR (EU) and CCPA (California), you have the right to access, correct, and delete your personal data. Enforcement of these rights may differ for services operated from outside your jurisdiction. Self-hosted Qwen deployments under Apache 2.0 give you direct data control independent of Alibaba Cloud infrastructure.
The EU AI Act classifies general-purpose AI models above certain capability thresholds under transparency and risk obligations. Qwen models deployed within the EU are subject to these provisions. Open-weight releases also carry downstream compliance responsibilities for the deploying organization under the EU AI Act's provider liability framework.
This publication is editorially independent. AI tool coverage reflects independent research, verified benchmarks, and editorial judgment. Where affiliate links are present, they are clearly disclosed and do not influence conclusions.