---
title: "Qwen3 235B A22B Instruct API hosted in the EU"
description: "Run Qwen3 235B A22B Instruct (2507) on inference.eu: a 256K-context open model behind an OpenAI-compatible API, processed in EU data centres."
canonical: "https://inference.eu/models/qwen3-235b-a22b-instruct/"
language: "en"
---

Chat · Qwen (Alibaba Cloud)

# Run Qwen3 235B A22B Instruct in the EU through an OpenAI-compatible API

Qwen3 235B A22B Instruct is the July 2025 update of Alibaba's large Qwen3 mixture-of-experts model, tuned for direct answers without a thinking phase. On inference.eu you call it with the OpenAI or Anthropic request format, your requests are processed in EU data centres, and nothing you send is used for training.

[Contact sales](https://inference.eu/contact/)[Example request](#example)

1.  01235B · 22B
2.  02EU data centres
3.  03Apache 2.0

## What Qwen3 235B A22B Instruct is

Qwen3 235B A22B Instruct, released as version 2507, is a mixture-of-experts language model from the Qwen team at Alibaba Cloud. It has 235 billion parameters in total, of which 22 billion are active for each token, so it has the knowledge of a very large model while doing roughly the work of a 22B model per token.

It runs in non-thinking mode only: it answers directly and does not produce a reasoning block first. Its native context window is 262,144 tokens. Qwen describes the 2507 update as bringing better instruction following, broader long-tail knowledge across many languages and stronger long-context understanding. It is released under the Apache 2.0 licence.

## What it does well, and where it falls short

This model is a strong general-purpose choice for chat, writing, summarisation, translation and tool use where you want fast, direct answers rather than long chains of thought. The 256K context lets it read entire contract bundles, large codebases or long transcripts in one request, and its multilingual knowledge covers the main European languages well.

Because it never thinks before answering, it is weaker than reasoning models on competition-style maths, hard logic puzzles and long agentic coding tasks. It accepts text only, so images need a multimodal model such as Qwen3.8 27B. It is also older than the Qwen3.6 and Qwen3.8 releases, which match or beat it on many tasks at a much smaller size.

## Calling Qwen3 235B through the API

inference.eu uses the OpenAI request format, so the OpenAI SDKs for Python and JavaScript, LangChain, LlamaIndex and most agent frameworks work without changes. Point the base URL at the inference.eu API, use your inference.eu key, and set the model to `qwen3-235b-a22b-instruct`.

Streaming, tool calls and JSON output work as usual. There is no need to pass `enable_thinking=False`, because this version never produces thinking blocks. The Anthropic Messages format is also available, so tools written for Claude can use Qwen3 235B by changing the base URL, key and model name. Your prompts and parsing code stay as they are.

## Data residency for Qwen3 235B

Qwen is developed by Alibaba Cloud in China, but on inference.eu the model runs on infrastructure in the European Union. Your prompts and its responses are processed in EU data centres, they are not sent to Alibaba, and they are never used to train any model.

Every customer receives a data processing agreement under Article 28 of the GDPR together with our list of subprocessors, so you know who processes the data and where. Billing is in euros. The Apache 2.0 licence places no restriction on commercial use, so you can build Qwen3 235B into paid products without a separate agreement with Alibaba.

## Qwen3 235B A22B Instruct

Maker

Qwen (Alibaba Cloud)

Made in

China

Type

Chat

Parameters

235B

Active parameters

22B

Context window

262,144 tokens

Licence

[Apache 2.0](https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507/blob/main/LICENSE)

Released

July 2025

Model ID

qwen3-235b-a22b-instruct

Processing

EU data centres

Training on your data

Never

Source

[huggingface.co](https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507)

qwen3-235b-a22b-instruct · Python

```
import os
from openai import OpenAI

client = OpenAI(base_url="https://api.inference.eu/v1", api_key=os.environ["INFERENCE_EU_API_KEY"])

response = client.chat.completions.create(
    model="qwen3-235b-a22b-instruct",
    messages=[{"role": "user", "content": "Summarise this contract in five bullet points."}],
)
print(response.choices[0].message.content)
```

## What teams use it for

-   1

    ### Long-document analysis

    Read contract bundles, tender documents or technical manuals of several hundred pages in a single request and answer questions about them.

-   2

    ### Multilingual customer support

    Draft replies and summarise tickets across European languages with direct answers and no reasoning delay.

-   3

    ### Tool-using assistants

    Call internal APIs, search and databases through function calling in assistants that need quick, accurate answers.

## Qwen3 235B A22B Instruct: common questions

The questions buyers, developers and legal teams ask us most. Something missing? Ask us directly.

GDPR compliance depends on how you process data, not on the model alone. On inference.eu, Qwen3 235B requests are processed in EU data centres, not used for training and covered by a data processing agreement under Article 28 GDPR with a published subprocessor list. That gives you the contractual basis your data protection officer needs.

On inference.eu, the open weights run on EU infrastructure, so no data goes to Alibaba or to China. What remains is the model's own training: like any model it reflects choices made by its maker, including on politically sensitive topics. Test it on your own prompts before relying on it.

The native context window is 262,144 tokens, about 256K. The Qwen team describes methods to extend it towards one million tokens, but the native window is what the model was trained for. Prompt and response together must fit inside it.

Yes. Qwen3 235B A22B Instruct is released under the Apache 2.0 licence, which allows commercial use, modification and redistribution without fees or user limits. You only need to keep the licence and copyright notices if you redistribute the weights.

A22B means 22 billion active parameters. The model has 235 billion parameters in total, split into experts, and only a subset of about 22 billion is used for each token. That keeps generation much cheaper than a dense model of the same size.

No. The 2507 Instruct version is non-thinking only and answers directly. If you need step-by-step reasoning for maths, logic or complex agents, use a reasoning model such as Qwen3.8 27B with thinking enabled, DeepSeek V4 Flash or GLM-5.3.

In your OpenAI client, change the base URL to the inference.eu API and use your inference.eu key. Then set the model to \`qwen3-235b-a22b-instruct\`. Messages, tools and streaming work the same way, so most applications need no other change.

Qwen3 235B has more total parameters, a context window twice as long and generally stronger results in coding, maths and multilingual knowledge. Llama 3.3 70B is a smaller dense model with a long track record. Both are text-only and both answer without a thinking phase.

No, this model accepts text only. For screenshots, charts and scanned documents, use Qwen3.8 27B or Qwen3.6 27B, which include a vision encoder, or Gemma 4 31B. For pure document text extraction, DeepSeek-OCR 2 is a specialised option.

No. Requests to inference.eu are used only to produce your response. They are not used to train or fine-tune Qwen or any other model, and they are not shared with Alibaba Cloud, which only published the weights.

[Ask a question](https://inference.eu/contact/)

## Related

[All models](https://inference.eu/models/#directory)

-   [Llama 3.3 70B InstructRun Meta's Llama 3.3 70B Instruct on inference.eu: an OpenAI-compatible API with requests processed in EU data centres and billed in euros.Chat](https://inference.eu/models/llama-3-3-70b-instruct/)
-   [Qwen3.8 27BRun Qwen3.8 27B on inference.eu: a dense open model with image and video input, adjustable reasoning and 256K context, OpenAI-compatible, processed in the EU.Reasoning](https://inference.eu/models/qwen3-8-27b/)
-   [DeepSeek V4 FlashRun DeepSeek V4 Flash (0731) on inference.eu: a 1M-token open reasoning model behind an OpenAI-compatible API, processed in EU data centres, not in China.Reasoning](https://inference.eu/models/deepseek-v4-flash/)
-   [Mistral Small 4 119BRun Mistral Small 4 119B from French lab Mistral AI on inference.eu: reasoning, vision and code in one model, processed in EU data centres.Chat](https://inference.eu/models/mistral-small-4-119b/)

## Move your AI workloads to EU infrastructure

Tell us which models you use and roughly how much traffic you have. We reply within one working day with pricing and next steps.

-   Reply within one working day
-   Pricing for your volume
-   DPA ready to sign