Qwen on RunAPI
Use OpenAI-compatible Chat Completions at https://runapi.ai/v1 as the primary protocol.
Primary protocol recipe
Authenticate
Set OPENAI_API_KEY to a RunAPI API key and OPENAI_BASE_URL to https://runapi.ai/v1.
Send request
from openai import OpenAI
client = OpenAI(api_key="YOUR_RUNAPI_TOKEN", base_url="https://runapi.ai/v1")
response = client.chat.completions.create(
model="qwen3.6-27b",
messages=[{"role": "user", "content": "Review this function for bugs."}],
)
print(response.choices[0].message.content)
print(response.usage)
For long output, call the same method with stream=True and
stream_options={"include_usage": True}; consume every chunk through [DONE].
Verify result
Require final assistant content, a terminal finish_reason, and authoritative
usage. Thinking models may also return reasoning_content; use the final
answer in content. A stream is complete only after its terminal usage chunk
and [DONE].
Stop boundaries
Correct a rejected shape once using the structured error. Retry transport once
only before any response or Usage and when replay is safe. Record a terminal
error and stop without changing model or protocol. Send image input only on
Chat Completions and only to a model whose row below lists image input. Keep
qwen3-coder-next requests to text.
Compatibility protocols
Load compatibility protocols only when an existing client requires OpenAI Responses, Anthropic Messages, or Gemini contents.
Capability boundary
| Capability | Current behavior |
|---|---|
| Text history, sync, streaming | Supported through OpenAI-compatible Chat Completions |
| Image input | Chat Completions only, for the Qwen3.6, Qwen3.5, and Qwen3-VL models |
| Function tools | Chat Completions; verify each tool call before acting on it |
| OpenAI Responses, Anthropic Messages, Gemini contents | Text requests; image input is rejected |
| Usage | Preserve returned token fields; never infer missing details |
Supported models
| Model ID | Use when |
|---|---|
qwen3.6-27b |
Dense agentic coding and reasoning with image input |
qwen3.6-35b-a3b |
Efficient MoE coding and reasoning with image input |
qwen3.5-397b-a17b |
Largest Qwen3.5 open-weight model with image input |
qwen3.5-122b-a10b |
Qwen3.5 MoE for text and image reasoning |
qwen3.5-27b |
Dense Qwen3.5 text and image workloads |
qwen3.5-35b-a3b |
Low-cost Qwen3.5 MoE workloads |
qwen3-coder-next |
Text-only coding agents and repository work |
qwen3-vl-235b-a22b-thinking |
Image reasoning with always-on thinking |
qwen3-vl-235b-a22b-instruct |
Image understanding and OCR without thinking |