Qwen (Alibaba)
Qwen 3.5 — open-weight under Apache 2.0 with strong multilingual coverage.
Alibaba's Qwen 3.5 122B sits alongside DeepSeek as the global leader in 'intelligence-per-parameter,' shipped under a permissive Apache 2.0 license. Qwen's standout strengths are multilingual coverage (especially across CJK languages) and strong instruction-following at compact parameter counts — making it the natural choice for agents that need to operate across multiple regions or that ship inside customer infrastructure under unrestricted commercial terms.
Models
Qwen 3.5 122B
OpenHigh MMLU, Apache 2.0 license — full commercial freedom
Context Window
128K
Max Output
32K tokens
Input Price
Apache 2.0 — self-host
Input Types
Text, Images
Output Types
Text, Code, JSON
Qwen 3.5 32B
Sweet spot for on-prem agents under cost pressure
Context Window
128K
Max Output
16K tokens
Input Price
Apache 2.0 — self-host
Input Types
Text
Output Types
Text, Code
Qwen 3 VL
Multimodal vision-language for document understanding
Context Window
128K
Max Output
8K tokens
Input Price
Apache 2.0 — self-host
Input Types
Text, Images
Output Types
Text
Use Cases
Multilingual Global Agents
Strong CJK and EMEA-language coverage at no per-token cost — ideal for global customer-service agents.
Embedded On-Prem AI
Apache 2.0 license means Qwen can ship inside customer products and air-gapped environments without extra contracts.
Document Vision
Qwen 3 VL handles invoices, contracts, and diagrams across languages without a separate OCR pipeline.
Frontier-Lite at Scale
Same role as DeepSeek — high quality at near-zero marginal cost for agent fleets.
Why use Qwen (Alibaba) on Nagent?
Nagent adds enterprise orchestration, observability, and workflow automation on top of Qwen (Alibaba)'s raw model capabilities.
Apache 2.0 is the most permissive license in the open-weight category — no usage strings attached
Self-host alongside Mistral / DeepSeek in the same sovereign-AI cluster
Strong multilingual coverage cuts the need for per-region model swaps
Drop-in replacement for closed models in our agent runtime
How to access Qwen (Alibaba) on Nagent
Open Agent Studio
Navigate to Agent Studio in your Nagent workspace.
Select Qwen
Pick the parameter tier that matches your hardware and quality target.
Deploy On-Prem
Use our self-hosted-model setup wizard to point Nagent at a Qwen endpoint inside your own VPC.
Common questions about Qwen (Alibaba)
Real buyer and developer questions, answered. Click any item to expand.
Apache 2.0 vs Mistral's research license — does it matter?
Yes for commercial deployment. Apache 2.0 lets you ship Qwen inside customer products with no further licensing. Mistral Large 3 currently requires a research license, which adds friction for embedded/OEM scenarios.
How does Qwen handle non-English languages?
Strong CJK coverage and good EMEA-language quality. The right choice if your agent fleet operates across regions and you would otherwise need per-language model swaps to maintain quality.
Is the parameter count meaningful — 122B vs 32B?
Yes for compute planning: 122B needs ~244GB VRAM at FP16, 32B needs ~64GB. On output quality, 32B is the sweet spot for most enterprise agent workloads — the 122B step gives diminishing returns outside specialised tasks.
Can I fine-tune Qwen on Nagent?
Yes for self-hosted deployments. Bring a LoRA adapter or do a full fine-tune; Nagent's training infrastructure supports both Qwen base and instruct variants on shared GPU clusters.
Ready to use Qwen (Alibaba) inside your agents?
Get started in minutes — no API key management required.