llm-api-adapter-qwen

An optional Qwen organization package for LLM API Adapter, a Python SDK with one shared interface for calling LLM APIs.

Uses Model Studio's Frankfurt Global deployment through its Anthropic-compatible Messages API.

Installation

Install through the Core package extra:

pip install "llm-api-adapter[qwen]"

Direct installation is also supported when Core is managed separately:

pip install llm-api-adapter-qwen

Async requests need HTTPX:

pip install "llm-api-adapter[qwen,async]"

Synchronous requests use requests by default. To opt into HTTPX for sync chat() and stream_chat(), install "llm-api-adapter[qwen,httpx]" and pass transport="httpx".

Quick start

import os

from llm_api_adapter.models.messages.chat_message import UserMessage
from llm_api_adapter.universal_adapter import UniversalLLMAPIAdapter

adapter = UniversalLLMAPIAdapter(
    organization="qwen",
    model="qwen3.8-max",
    api_key=os.environ["QWEN_API_KEY"],
)

response = adapter.chat(
    messages=[UserMessage("Explain retrieval-augmented generation.")],
    max_tokens=128,
    workspace_id=os.environ["QWEN_WORKSPACE_ID"],
)
print(response.content)

The package supports only Model Studio's Frankfurt Global deployment. Pass the required workspace_id explicitly to every chat, stream_chat, achat, and astream_chat call; it is never read from an environment variable. The package uses:

https://{workspace_id}.eu-central-1.maas.aliyuncs.com/apps/anthropic/v1/messages

Supported models

• qwen3.8-max
• qwen3.8-flash
• qwen3.7-plus
• qwen3.7-flash

Capabilities

All four models default to hybrid thinking. Set reasoning_level="none" to disable thinking, or capture_reasoning=True to receive provider-emitted reasoning separately from visible text. max_tokens must be a positive integer and limits generated output; it is separate from Qwen 3.7's thinking budget. With thinking enabled, Model Studio's reported usage.output_tokens can also include thinking tokens, so it can exceed max_tokens even when the visible answer respects that output limit.

Qwen permits tool_choice="auto" and "none" in thinking mode, but not a forced "any" or named tool. For a forced tool call, the adapter automatically disables thinking and issues a UserWarning; pass reasoning_level="none" to make that choice explicit without a warning.

Both forced tool-choice modes are declared pass exceptions: the adapter handles the provider restriction while preserving the tool-call contract. previous_response is accepted but ignored; supply complete messages history, including assistant tool calls and tool results, on each turn. Exact rules and exceptions are recorded in the Qwen registry.

PDF input

The package rejects DocumentPart URLs and bytes before any HTTP request. It does not upload documents or run OCR.

Automatic cache usage and pricing

Model Studio's Messages usage reports ordinary input_tokens separately from cache_read_input_tokens. When both counts are present, the adapter adds them to normalize total Usage.input_tokens and exposes cache reads as Usage.cached_tokens. The registry tier is selected from this total, then ordinary and cached input use their own Frankfurt Global CNY rates.

No automatic cache-write component is priced, and cache_write_tokens remains None. An omitted cache-read split leaves cost_input and cost_total unknown; known output cost can remain available. Opt-in caching, TTLs, storage, batch, promotional, and negotiated rates are excluded. See the shared usage and pricing guide.

See the main llm-api-adapter README for the shared API contract and examples.