Skip to main content

​
Claude Thinking Models

Claude models with the -think suffix enable enhanced reasoning capabilities, suitable for complex reasoning, planning, and multi-step analysis.

​
Supported Models

​
Notes

  1. Thinking capability is explicitly selected via the model name
  2. Claude thinking models use the platform’s default context and token configuration
    • Sonnet series: default max_tokens = 32k
    • Opus series: default max_tokens = 64k
  3. No additional parameters are required; invocation is identical to standard models

​
GPT Reasoning Models

GPT models support different reasoning intensities via suffixes, allowing a balance between output quality, latency, and cost.

​
Supported Models

​
Notes

  1. -high indicates high-intensity reasoning, suitable for complex tasks requiring higher accuracy
  2. Reasoning intensity is determined by the model name; no extra fields are required

​
Google Search–Enhanced Models

Selected Gemini models integrate Google’s official search capability.
By choosing the appropriate model name, search is automatically enabled during generation—no additional parameters required.

​
Supported Models

​
Notes

  1. Models with the -search suffix include Google’s official search integration and are suitable for real-time information retrieval, external fact-checking, and up-to-date references
  2. Search incurs additional costs, which are recorded as a separate usage log and included in the total charges
    • Detailed search cost breakdowns are not currently displayed and will be added in a future update
  3. Only OpenAI-compatible API formats are supported
    • Gemini native SDK is not supported for these models
    • If using the Gemini official SDK, please refer to the corresponding non-search model examples and configure search using Google’s official parameters