xAI

Route your xAI API calls through AI SpendOps for automatic usage tracking and cost attribution.

Configuration

Setting	Value
Route	/v1/xai/*
Upstream	https://api.x.ai
Auth header	Authorization: Bearer ...
Streaming usage	Auto-injected (stream_options.include_usage)

SDK base URL

https://proxy.aispendops.com/v1/xai/v1

Example

curl https://proxy.aispendops.com/v1/xai/v1/chat/completions \
  -H "Authorization: Bearer your-xai-key" \
  -H "X-ASO-API-Key: aso_k_yourkey.secret" \
  -H "Content-Type: application/json" \
  -d '{"model":"grok-3","messages":[{"role":"user","content":"Hello"}]}'

Python SDK

xAI is OpenAI-compatible, so the OpenAI SDK works with a changed base URL:

from openai import OpenAI

client = OpenAI(
    api_key="your-xai-key",
    base_url="https://proxy.aispendops.com/v1/xai/v1",
    default_headers={"X-ASO-API-Key": "aso_k_yourkey.secret"},
)

response = client.chat.completions.create(
    model="grok-3",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Usage fields

Field	Description
prompt_tokens	Input tokens, including any cached tokens
completion_tokens	Output tokens, including any reasoning tokens
total_tokens	Sum of prompt and completion tokens
prompt_tokens_details.cached_tokens	Portion of prompt_tokens served from cache
completion_tokens_details.reasoning_tokens	Portion of completion_tokens spent on reasoning

xAI follows the OpenAI usage shape, so cached tokens are reported inclusively as a breakdown inside prompt_tokens. AI SpendOps subtracts the cached portion before pricing so it is charged once at the cache-read rate. See OpenAI for the full explanation, and Anthropic for the provider that does the opposite.

Watching reasoning spend on Grok

Grok's reasoning models bill thinking at the output rate, and that thinking never appears in the response. A request can return three sentences and be charged for thousands of output tokens.

Because AI SpendOps records reasoning_tokens as its own field on every event, you can see the split directly rather than inferring it from an invoice total:

• Compare reasoning_tokens against completion_tokens to get the share of output spend that is thinking rather than answers.
• Break that down by dimension to find which feature or team is driving it.
• If a reasoning model is being used where a standard model would do, model allow-lists can stop a given ASO key reaching it at all.

What AI SpendOps records

Field	Populated
cache_read_tokens	Yes, from prompt_tokens_details.cached_tokens
cache_write_tokens	No. There is no separate cache write charge
reasoning_tokens	Yes for reasoning models
provider_cost	No. Cost is calculated from token counts and the current xAI rate card

Notes

• The proxy injects stream_options: { include_usage: true } on streaming requests. See Streaming.
• Full field definitions are in the usage event field reference.