Use priority processing

Add service_tier: "priority" on a Chat Completions or Responses request body for higher scheduling priority (typically lower TTFT and faster inter-token latency under load). Values are "default" (same as omitting the field) or "priority".

The response includes service_tier confirming which tier ran. You are only billed at the priority rate when the response says "priority". Priority is a premium per-token rate; cache discounts apply before the multiplier.

Responses curl

curl https://api.x.ai/v1/responses \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-4.6",
    "input": "Explain the Riemann hypothesis in one paragraph.",
    "service_tier": "priority"
  }'

OpenAI SDK

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.getenv("XAI_API_KEY"),
    base_url="https://api.x.ai/v1",
)

response = client.responses.create(
    model="grok-4.6",
    input="Explain the Riemann hypothesis in one paragraph.",
    service_tier="priority",
)

print(response.output_text)
print(f"Tier used: {response.service_tier}")

xAI SDK

import os
from xai_sdk import Client
from xai_sdk.chat import user

client = Client(api_key=os.getenv("XAI_API_KEY"))

chat = client.chat.create(
    model="grok-4.6",
    service_tier="priority",
)
chat.append(user("Explain the Riemann hypothesis in one paragraph."))

response = chat.sample()
print(response.content)
print(f"Tier used: {response.service_tier}")

Best for latency-sensitive user-facing paths. Background, eval, and bulk work fit the Batch API better.

Pitfalls

• Check service_tier on every response. Priority capacity is not guaranteed; a "default" reply means default billing.
• Log the returned tier next to your latency metrics so you can see how often priority actually lands.
• Combine with prompt caching. Cached input tokens are discounted before the priority multiplier.
• Console API credits are separate from SuperGrok's weekly pool on grok.com.