← Back to Blog
New model

Qwen3.8 27B hosted in Japan: japan-qwen3.8-27b API

FastMetal

Qwen3.8 27B now has a second route on FastMetal whose inference runs inside Japan. The model id is japan-qwen3.8-27b, and it works through the API and in chat.

The existing qwen3.8-27b stays as it is. Both routes serve the same model; what differs is where inference happens and which inputs each route accepts. When a project requires processing in Japan, call japan-qwen3.8-27b; otherwise keep qwen3.8-27b. Switching is a change of model name.

Capabilities

  • Hosted and run in Japan. Requests are processed on servers inside Japan. In the catalog and the chat it carries a Japan badge, like every other Japan-hosted route.
  • 1M-token context, for feeding long documents or a codebase in one request.
  • Image input. We checked it with real images before launch.
  • Tool calling and structured output. Function calling (tools) and strict json_schema responses both worked on real requests.
  • Prompt caching. Send the same prefix again and the repeated part bills at the cached-input rate. Agents with long system prompts and tool definitions benefit most.
  • A reasoning model. It thinks before answering, and those tokens bill as output. Set reasoning_effort to none to skip the thinking.

How prompts are retained and whether they are used for training is listed on the model page. Check it before sending sensitive data.

Pricing

Billed per million input and output tokens, with cached input at its own rate. Billing is prepaid, so you can never spend more than you have loaded. Current rates are on the model page and pricing.

Calling it

FastMetal's API is OpenAI-compatible. Set the model to japan-qwen3.8-27b:

curl https://api.fastmetal.ai/v1/chat/completions \
  -H "Authorization: Bearer $FASTMETAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "japan-qwen3.8-27b",
    "messages": [
      {"role": "user", "content": "List the decisions and owners from these meeting notes as a table."}
    ]
  }'

Already on qwen3.8-27b? Changing the model name is the whole migration. To try it without code, pick the Japan-hosted Qwen3.8 27B in chat.

Before you build on it

  • No video or PDF input. This route accepts text and images; a video or PDF part returns 400.
  • Four reasoning levels. reasoning_effort takes none, low, medium and xhigh; high and max are mapped to the nearest supported level.
  • auto never picks it. It is not a candidate for model: "auto"; it serves only requests that name it.
  • Expect an occasional 429 at busy times. Japan-hosted capacity is limited, and when it is full the route returns 429. Retry after a short wait.

Real answers from the same model to our shared prompts are on the Qwen3.8 27B model page. For the other Japan-hosted models, look for the Japan badge in the model catalog.

Try Qwen3.8 27B (Japan) right now

Qwen3.8 27B (Japan) is available on FastMetal through one API key. Start in the browser, or call it from the OpenAI SDK.