TOKI
Models

Model Parameters

Learn how to choose a model and set common parameters in OpenAI-compatible requests.

Specify A Model

For chat completions, model is required. Use the model identifier shown on the model page or in the console:

const completion = await openai.chat.completions.create({
  model: 'deepseek/deepseek-chat-v3',
  messages: [{ role: 'user', content: 'Explain quantum computing' }],
});

Common Parameters

ParameterDescription
temperatureControls randomness. Higher values make output more varied.
top_pNucleus sampling parameter. Avoid changing it aggressively together with temperature.
max_tokensLimits how many tokens the response can generate.
streamSet to true to receive an SSE stream.
response_formatRequests a specific output format such as JSON; support depends on the model.

Capability Differences

Support can differ by model for:

  • Context length
  • Image or multimodal input
  • Tool calls
  • JSON mode
  • Reasoning fields
  • Sampling parameters

If a parameter is not supported by the selected model, the server may return a parameter error or ignore some optional parameters. For production, validate the smallest request shape with the target model first.

Selection Suggestions

  1. Start with the required capability, such as text, image, long context, or tool calling.
  2. Then compare response speed, price, and stability.
  3. For production tasks, use separate API keys and quota limits for different scenarios to simplify monitoring and risk control.

On this page