# Grok

> Call Grok 4.6 through RunAPI Responses only; use Grok 4.3, 4.5, or Grok 4.20 non-reasoning through their verified OpenAI-compatible interfaces. Use for text, streaming, tools, structured output, or an existing compatibility client that needs the conditional reference.

- Skill: `runapi-ai/grok` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add runapi-ai/grok`
- Raw SKILL.md: https://api.skillmd.com/api/skills/runapi-ai/grok/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: runapi-ai (https://skillmd.com/u/runapi-ai)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/runapi-ai/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

```python
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](references/compatibility-protocols.md) only when an existing client requires Grok 4.3 or 4.5 through Anthropic Messages or Gemini contents.

## Supported models

| Model ID | Use when |
|---|---|
| `grok-4.20-0309-non-reasoning` | Verified text workloads without reasoning controls |
| `grok-4.6` | Responses workflows with image input, function tools, structured output, and low through xhigh reasoning |
| `grok-4.5` | Current Grok chat and reasoning workloads |
| `grok-4.3` | Stable Grok reasoning workloads |

## References

- <https://runapi.ai/models/grok.md>
- <https://runapi.ai/providers/xai.md>
- <https://runapi.ai/models.md>