grok

Grok on RunAPI

Use OpenAI-compatible endpoints 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.responses.create(
    model="grok-4.6",
    input=[{
        "role": "user",
        "content": [{"type": "input_text", "text": "Review this rollout plan."}],
    }],
    stream=False,
    reasoning={"effort": "xhigh"},
)
print(response.output_text)
print(response.usage)

Use Chat Completions for grok-4.5 chat workflows. Use Responses for grok-4.3, grok-4.5, or grok-4.6; Grok 4.6 does not expose the other compatibility protocols. Grok 4.6 accepts function tools and image input; add an input_image part with a public image_url alongside input_text for image understanding. Only function tools are supported; hosted tools such as web_search, x_search, file_search, code_interpreter, image_generation, and mcp are rejected. Do not send input_file or state fields such as previous_response_id for Grok 4.6. For streaming Responses, set stream=True and consume through response.completed, terminal usage, and [DONE].

Verify result

Responses require final output, one usage-bearing response.completed, and [DONE]. Chat requires final assistant content, finish_reason, and terminal usage.

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. Keep grok-4.20-0309-non-reasoning text-only and stateless; add advanced controls only when the current RunAPI contract verifies them for the exact model.

Compatibility protocols

Load compatibility protocols only when an existing client requires Grok 4.3 or 4.5 through Anthropic Messages or Gemini contents.

Supported models

References