# Grok API: model IDs, reasoning effort, 2M context, and errors

Source: https://ofox.ai/docs/develop/models/grok

xAI's Grok models, including the reasoning models Grok 4.7, 4.6, 4.5, and 4.3, and the non-reasoning models Grok 4.1 Fast and 4.20. Called through the OpenAI-compatible API.

## Common tasks

- [Reduce reasoning time](#reasoning) — What reasoning_effort does and its measured effect.
- [Endpoint support](#endpoints) — Chat Completions, Responses, and Anthropic endpoints.
- [Troubleshooting errors](#errors) — Reasoning parameter not supported, protocol not supported, request timeouts.

## Models
For the full model list and live prices, see the [models page](https://ofox.ai/models?provider=x-ai). Model IDs have the prefix `x-ai/`, which can be omitted.

**Common model IDs (copy them exactly)**

- Grok 4.7: `x-ai/grok-4.7` — Reasoning model, 500K context
- Grok 4.1 Fast: `x-ai/grok-4.1-fast` — Non-reasoning, 2M context, image support

| Type | Models | Description |
|---|---|---|
| Reasoning models | Grok 4.7, 4.6, 4.5 | Always reason and take longer; 500K context, text only |
| Reasoning model | Grok 4.3 | 1M context, supports image input |
| Non-reasoning models | Grok 4.1 Fast, 4.20 | Fast responses; 2M context, supports image input |

## Endpoint support
| Endpoint | Supported models |
|---|---|
| Chat Completions `/v1/chat/completions` | All |
| Responses `/v1/responses` | Grok 4.3, 4.20, 4.1 Fast |
| Anthropic native API | Grok 4.3 only; others return `protocol_not_supported` |

Grok models are provided by xAI directly (`grok`), and some models also have Azure AI Foundry as a provider.

## Examples

**Python**

```python filename="grok_basic.py"
import os
from openai import OpenAI

client = OpenAI(base_url="https://api.ofox.ai/v1", api_key=os.environ["OFOX_API_KEY"])

resp = client.chat.completions.create(
    model="x-ai/grok-4.7",
    reasoning_effort="low",
    messages=[{"role": "user", "content": "Explain idempotency in one sentence."}],
)
print(resp.choices[0].message.content)
print("Reasoning tokens:", resp.usage.completion_tokens_details.reasoning_tokens)
```

**cURL**

```bash filename="Terminal"
curl https://api.ofox.ai/v1/chat/completions \
  -H "Authorization: Bearer $OFOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "x-ai/grok-4.7",
    "reasoning_effort": "low",
    "messages": [{"role": "user", "content": "Explain idempotency in one sentence."}]
  }'
```

## Reasoning

**Key point**:Grok 4.7, 4.6, and 4.5 always reason, and complex questions may take several minutes. Use reasoning_effort: low to reduce the amount of reasoning, or switch to the non-reasoning Grok 4.1 Fast.

| Model | Reasoning | `reasoning_effort` |
|---|---|---|
| Grok 4.7, 4.6, 4.5 | Always reason | `low` significantly reduces reasoning |
| Grok 4.3 | Always reasons; reasoning content is not returned | Accepted |
| Grok 4.1 Fast, 4.20 | No reasoning | Returns 400 on Grok 4.1 Fast |

Measured on the same math problem: by default Grok 4.7 used about 15,000 reasoning tokens and took about 4 minutes; with `reasoning_effort: "low"`, it used about 4,200 tokens and took about 1 minute. Reasoning tokens are billed at the output rate, and their count is in `usage.completion_tokens_details.reasoning_tokens`; reasoning content is in `delta.reasoning_content` of the streaming response.

## Other features
- **Image input**: supported by Grok 4.3, 4.20, and 4.1 Fast, using standard `image_url` content blocks; see [Vision](https://ofox.ai/docs/develop/guides/vision) for the format.
- **Structured output and tool calling**: `response_format` supports `json_schema`; tool calling uses the standard `tools` parameter.
- **Caching**: repeated request prefixes are cached automatically; the hit count is in `usage.prompt_tokens_details.cached_tokens`.

## Common errors
| HTTP · type | Error message | Cause | Fix |
|---|---|---|---|
| 400 `invalid_request_error` | Contains `This model does not support` and `reasoning` | `reasoning_effort` was sent to the non-reasoning model Grok 4.1 Fast | Remove the parameter |
| 400 `protocol_not_supported` | `Model 'x-ai/grok-4.7' does not support the 'anthropic' protocol on this platform.` | The model does not support the Anthropic API | Use the OpenAI-compatible API |
| 400 `endpoint_not_supported` | `Model 'x-ai/grok-4.7' does not support the /v1/responses endpoint on this platform.` | The model does not support the Responses API | Use `/v1/chat/completions` |
| 404 `model_not_found` | `Model 'xai/grok-4.7' not found` | The prefix was written as `xai/` | The prefix is `x-ai/` |
| Request timeout | — | Reasoning takes too long | Pass `reasoning_effort: "low"`, use streaming, or set the client timeout to 600 seconds or more |

For other general errors, see [Error handling](https://ofox.ai/docs/develop/guides/error-handling).

## Official docs
- [xAI · Documentation](https://docs.x.ai/)

## FAQ

**How do I write Grok model IDs?**

The standard form is x-ai/grok-4.7; you can also omit the prefix and write grok-4.7. Writing xai/grok-4.7 returns 404.

**Why is Grok 4.7 so slow?**

Grok 4.7, 4.6, and 4.5 always reason, and complex questions may use more than ten thousand reasoning tokens and take several minutes. Passing reasoning_effort: "low" shortens this significantly, or switch to the non-reasoning Grok 4.1 Fast.

**Can reasoning be turned off for Grok?**

Reasoning cannot be turned off for Grok 4.7, 4.6, 4.5, or 4.3; you can only reduce it with reasoning_effort. For fast responses, use Grok 4.1 Fast.

**Can Claude Code use Grok?**

Only Grok 4.3 supports the Anthropic API; calls to other Grok models return protocol_not_supported.

**How long a context does Grok support?**

Grok 4.1 Fast and 4.20 support 2M tokens, Grok 4.3 supports 1M, and Grok 4.7, 4.6, and 4.5 support 500K.