Generative AI Service¶

This system provides a generative AI service based on large language models (LLMs). The API is OpenAI-compatible, so you can use it from your own machine or applications by simply switching the base URL of the OpenAI SDK or a compatible tool.

Available Models¶

Parameters are total parameter counts. All models except qwen3.8-27b are MoE (Mixture of Experts) models, so the number of parameters actually used to generate a token is smaller. Context length is the maximum number of input tokens accepted by this service, which may differ from the model's native limit.

All models support reasoning and function calling.

In addition to text, qwen3.8-27b, glm-5.3-flash, and deepseek-v4.1-flash also accept image input. The other models are text-only.

Creating an API Key¶

Create an API key in the RIKYU Portal.

RIKYU Portal

Log in to the portal and select API Keys at the top of the page. In the Create a new key form, choose a project ID and click CREATE.

One key can be created per project. API usage is billed to the project the key was created for. If you belong to multiple projects, use the key for the project you want to use.

Using curl¶

Set your API key in an environment variable. OPENAI_API_KEY is the variable name that the OpenAI SDK and many compatible tools read by default.

export OPENAI_API_KEY=API_KEY

Retrieve the list of available models.

curl https://api.rikyu.r-ccs.riken.jp/v1/models \
    -H "Authorization: Bearer $OPENAI_API_KEY"

Generate text. Specify the model you want to use in model.

curl https://api.rikyu.r-ccs.riken.jp/v1/chat/completions \
    -H "Authorization: Bearer $OPENAI_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "qwen3.6-35b",
      "messages": [
        {"role": "user", "content": "What is a supercomputer?"}
      ]
    }'

To receive the generated text incrementally, set stream to true.

curl https://api.rikyu.r-ccs.riken.jp/v1/chat/completions \
    -H "Authorization: Bearer $OPENAI_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "qwen3.6-35b",
      "messages": [
        {"role": "user", "content": "What is a supercomputer?"}
      ],
      "stream": true
    }'

Using Python¶

Install the OpenAI Python library.

pip install openai

Specify the endpoint in base_url and your API key in api_key.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.rikyu.r-ccs.riken.jp/v1",
    api_key=os.environ["OPENAI_API_KEY"],
)

response = client.chat.completions.create(
    model="qwen3.6-35b",
    messages=[
        {"role": "user", "content": "What is a supercomputer?"},
    ],
)

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

To receive the generated text incrementally, specify stream=True.

stream = client.chat.completions.create(
    model="qwen3.6-35b",
    messages=[
        {"role": "user", "content": "What is a supercomputer?"},
    ],
    stream=True,
)

for chunk in stream:
    content = chunk.choices[0].delta.content
    if content:
        print(content, end="", flush=True)

Using OpenAI-Compatible Applications¶

Any application that lets you configure the base URL and an API key, such as Cline (a Visual Studio Code extension), works as is. Configure it as follows.

The names of the settings differ between applications, but any application that lets you configure the base URL, an API key, and a model name can be used in the same way.

Notes¶

• If you do not specify max_tokens, the maximum number of tokens generated per request is 32,768. When you send a long input, adjust max_tokens so that the input plus max_tokens does not exceed the context length of the model. Exceeding it results in an error (HTTP 400).
• The maximum request body size is 32 MB.
• The request timeout is 600 seconds.
• Only paths under /v1 are publicly available.

Usage Fees¶

The generative AI service is free of charge during Early Access Phase 2.

The usage fee described in Welcome (300 JPY per GPU hour) applies to compute jobs run with Slurm; the generative AI service is not subject to it. You can check your API usage on the usage page of the RIKYU Portal.