[Alibaba Cloud](/ai-gateway/models/labs/alibaba)

# Qwen 3 32B

Qwen 3 32B is a dense 32-billion-parameter model from Alibaba Cloud with context of 128K tokens and hybrid thinking modes, reaching performance levels previously associated with much larger models. Your use is subject to Alibaba Cloud's [Terms](https://www.alibabacloud.com/help/en/legal/latest/alibaba-cloud-international-website-product-terms-of-service-v-3-8-0) & [Privacy](https://www.alibabacloud.com/help/en/legal/latest/alibaba-cloud-international-website-privacy-policy) Policies.

ReasoningTool Use

[Use with AI Gateway](https://vercel.com/d?to=%2F%5Bteam%5D%2F%7E%2Fai%3Futm_source%3Dgateway-model-page%26utm_campaign%3Dai-gateway-models&title=Get+Started+with+Vercel+AI+Gateway) [View docs](https://vercel.com/docs/ai-gateway)

AI SDKChat CompletionsMessagesResponses

```
1import { streamText } from 'ai'
2

3const result = streamText({
4  model: 'alibaba/qwen-3-32b',
5  prompt: 'Why is the sky blue?'
6})
```

[Read docs](https://vercel.com/docs/ai-gateway/sdks-and-apis/ai-sdk)

[Overview](/ai-gateway/models/qwen-3-32b) [API](/ai-gateway/models/qwen-3-32b/api) [About](/ai-gateway/models/qwen-3-32b/about) [Providers](/ai-gateway/models/qwen-3-32b/providers) [Throughput](/ai-gateway/models/qwen-3-32b/throughput) [Latency](/ai-gateway/models/qwen-3-32b/latency) [Uptime](/ai-gateway/models/qwen-3-32b/uptime) [Status](/ai-gateway/models/qwen-3-32b/status) [Similar](/ai-gateway/models/qwen-3-32b/similar) [FAQ](/ai-gateway/models/qwen-3-32b/faq)

## [Copy link to heading](#about-qwen-3-32b)About Qwen 3 32B

Qwen 3 32B is a fully dense model with no expert routing or sparse activation. All 32 billion parameters participate in generating each token. This architecture has a predictable operational profile: memory requirements are fixed, throughput is predictable, and there's no MoE infrastructure complexity to manage.

Alibaba Cloud positions Qwen 3 32B as reaching capability levels that Qwen2.5 required 72 billion parameters to achieve, a meaningful efficiency gain at the same parameter count from the third-generation architecture refinements across 64 transformer layers.

Hybrid thinking mode is available here as in the rest of the Qwen3 family. Activating thinking mode enables Qwen 3 32B to reason step-by-step before producing its answer, improving quality on problems requiring multi-step logic or structured derivation. Non-thinking mode bypasses the reasoning trace for applications where response speed takes priority. The budget control mechanism lets you set a token ceiling on the thinking phase, giving fine-grained control over the latency-quality tradeoff per request.

The model supports tool calling, agentic task scenarios, and MCP. The context window of 128K tokens accommodates long documents, multi-turn conversations, and retrieval-augmented generation (RAG) patterns where large amounts of source material need to fit in a single context.