Text Generation Models

Qwen language models for chat, reasoning, and code — from ultra-fast to flagship.

Text Generation Models

QwenAPI offers the full Qwen model family for text generation, ranging from ultra-low-latency models to flagship reasoning models with extended context windows.

Model Overview

Model	Context	Input	Output	Notes
`qwen3-max`	262K	$1.20/1M	$6.00/1M	Flagship, thinking mode
`qwen3.5-plus`	1M	$0.40/1M	$2.40/1M	Balanced, multimodal
`qwen3.5-flash`	1M	$0.10/1M	$0.40/1M	Fast and affordable
`qwen-turbo`	1M	$0.05/1M	$0.20/1M	Ultra-fast, lowest cost
`qwen3-coder-plus`	1M	$0.40/1M	$2.40/1M	Code specialized
`qwen3-coder-flash`	1M	$0.10/1M	$0.40/1M	Fast code model

Open Source Models

Self-hosted or API access for open-weight models:

qwen3.5-397b-a17b — 397B MoE, top open-source performance
qwen3-235b-a22b — 235B MoE flagship open model
qwen3-32b — Dense 32B, strong reasoning

Quickstart

``python


from openai import OpenAI

client = OpenAI(
    api_key="YOUR_QWENAPI_KEY",
    base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
)

response = client.chat.completions.create(
    model="qwen3-max",
    messages=[{"role": "user", "content": "Explain transformer attention in one paragraph."}],
)

print(response.choices[0].message.content)



Thinking Mode

qwen3-max supports extended chain-of-thought reasoning. Enable it with the enable_thinking extra parameter. The model returns a block before its final answer.

`python


response = client.chat.completions.create(
    model="qwen3-max",
    messages=[{"role": "user", "content": "What is 17 * 83?"}],
    extra_body={"enable_thinking": True},
)

Thinking content is in the first choice's message
print(response.choices[0].message.content)



Thinking mode increases latency and token usage. Use it for math, logic, and multi-step reasoning tasks where accuracy matters more than speed.

Web Search

Any model supports grounded responses via live web search:

`python


response = client.chat.completions.create(
    model="qwen3.5-plus",
    messages=[{"role": "user", "content": "What happened in AI news this week?"}],
    extra_body={"enable_search": True},
)



Choosing a Model

Use qwen-turbo for high-volume, latency-sensitive tasks: classification, extraction, simple Q&A.

Use qwen3.5-flash when you need a step up in quality without a big cost increase.

Use qwen3.5-plus for general-purpose assistants, RAG pipelines, and multimodal inputs.

Use qwen3-max for complex reasoning, agentic workflows, and tasks where output quality is critical.

Use qwen3-coder-plus/flash for code generation, review, and debugging.



Model Versioning

Model names without a date suffix (e.g., qwen3-max) always point to the latest stable version. To pin a specific version, append the release date:


qwen3-max-2025-09-19



Pinned versions are supported for at least 6 months after a new stable release.

Batch and Caching

qwen3-max` supports both the Batch API (50% off) and prompt caching. Cached tokens are billed at 10% of the standard input rate.

See Batch API for async bulk processing.