# Qwen 3.6 27B

Qwen 3.6 27B is a hybrid-architecture native vision-language model from Alibaba Cloud in the Qwen 3.6 line, tuned for agentic coding, mathematical and code reasoning, and spatial intelligence over a context window of 256K tokens.

- **Model ID:** `alibaba/qwen3.6-27b`
- **Type:** chat
- **Providers:** alibaba
- **Context window:** 256,000
- **Maximum output tokens:** 256,000
- **Pricing:** $0.6/1M input tokens, $3.6/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/qwen3.6-27b

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'alibaba/qwen3.6-27b',
  prompt: 'Why is the sky blue?'
})
```

## About

Qwen 3.6 27B, released on April 22, 2026, is a Qwen 3.6 native vision-language model in Alibaba Cloud's Qwen 3 family. It is built on a hybrid architecture that combines linear attention mechanisms with a sparse mixture-of-experts (MoE) framework, a design intended to keep inference efficient at long context while preserving the capability profile of a larger network.

Compared with the prior 3.5-35B-A3B generation, Qwen 3.6 27B brings improvements across several axes. Agentic coding ability is stronger, which matters for pipelines that chain tool calls and multi-step plans. Mathematical and code reasoning have been upgraded for benchmark-style problem solving and real-world programming tasks. Spatial intelligence, object localization, and object detection are sharper, which improves the model's grounding when it must reason about positions of elements within an image.

Native vision-language support means images are treated as first-class inputs alongside text rather than processed through a bolt-on encoder. Qwen 3.6 27B accepts file input and is tagged for reasoning, tool use, and implicit caching, so it can ingest documents and images, decide when to invoke registered tools, and reuse cached prefixes when serving repeated long prompts. The context window of 256K tokens accommodates extended multimodal sessions, document plus image inputs, and long agent traces.

You can integrate Qwen 3.6 27B through AI SDK, Chat Completions API, Responses API, Messages API, or other API formats, from TypeScript or Python. Maximum output is 256K tokens per request.

## What to consider

Sparse MoE architectures keep active compute small per token, but providers serve the model through different infrastructure paths. Check the live cost and latency metrics on this page before sizing high-volume workloads, and confirm vision payload limits with your selected provider.

## When to use

### Best For

- Agentic coding pipelines that combine reasoning, tool use, and image input within a single multimodal session
- Vision-language workloads that require sharper object localization, detection, and spatial grounding
- Mathematical and code reasoning tasks where step-by-step accuracy matters more than raw throughput
- Long-context multimodal applications that benefit from the window of 256K tokens for combined text and image inputs
- Workloads where implicit caching reduces cost on repeated long prefixes such as large system prompts

### Consider Alternatives When

- Workloads are text-only and never use vision, where a dedicated text model offers lower cost per token
- Extended chain-of-thought matters most, in which case a thinking-mode model is a better match
- Latency-critical pipelines need a smaller, faster vision model rather than this capability tier
- Image or video generation is required rather than understanding, which calls for a generation-class model

## Best for

- **Agentic Multimodal Coding:** Pipelines that combine reasoning, tool use, and image input within a single session
- **Spatial And Object Tasks:** Workloads requiring sharper object localization, detection, and 2D grounding
- **Math And Code Reasoning:** Step-by-step accuracy in mathematical and programming problems over raw throughput
- **Long-Context Multimodal Sessions:** Combined text and image inputs handled within the window of 256K tokens
- **Repeated Long Prefixes:** Implicit caching reduces cost on shared system prompts and document inputs

## Consider alternatives

- **Text-Only Workloads:** A dedicated text model offers lower cost per token when vision is never used
- **Maximum Reasoning Depth:** A thinking-mode model is a better match when extended chain-of-thought matters most
- **Latency-Critical Pipelines:** A smaller, faster vision model serves simple multimodal tasks at lower cost
- **Image Or Video Generation:** A generation-class model fits tasks that produce pixels rather than read them

## Frequently asked questions

### What architecture does Qwen 3.6 27B use?

Qwen 3.6 27B is built on a hybrid architecture that integrates linear attention mechanisms with a sparse mixture-of-experts framework. The combination is designed to keep inference efficient at long context while preserving the capability profile of a larger network.

### How does Qwen 3.6 27B compare to the prior Qwen 3.5-35B-A3B generation?

Alibaba Cloud positions Qwen 3.6 27B as a clear step up from the 3.5-35B-A3B generation, with significantly improved agentic coding, mathematical and code reasoning, spatial intelligence, and object localization and detection performance.

### What modalities does Qwen 3.6 27B accept?

Qwen 3.6 27B is a native vision-language model. It accepts interleaved text and images within a single context window of up to 256K tokens, and supports file input alongside text and vision.

### Does Qwen 3.6 27B support tool calling?

Yes. Qwen 3.6 27B is tagged for tool use and is tuned for agentic coding workflows. The model can select and invoke registered functions across multi-turn sessions through AI Gateway.

### How do I access Qwen 3.6 27B through AI Gateway?

Authenticate with an AI Gateway API key or OIDC token and reference ``alibaba/qwen3.6-27b`` as the model. You can call Qwen 3.6 27B through AI SDK, Chat Completions API, Responses API, Messages API, or other API formats, from TypeScript or Python.

### Does Qwen 3.6 27B support zero data retention?

Yes, Zero Data Retention is available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

### Where can I see live latency and cost data for Qwen 3.6 27B?

This page shows live throughput, time-to-first-token, and pricing metrics for Qwen 3.6 27B measured across real AI Gateway traffic.

## Links

- [Model page](https://vercel.com/ai-gateway/models/qwen3.6-27b)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://modelstudio.console.alibabacloud.com/ap-southeast-1?spm=a2ty_o05.31384571.0.0.56899f6bACyZax&tab=doc#/doc/?type=model&url=2840914_2&modelId=qwen3.6-35b-a3b&serviceSite=international)


If you encounter incorrect information, can’t find what you need, need a capability Vercel doesn’t currently offer, or have any other feedback to report, please do it by following the instructions at https://vercel.com/feedback.md.