Flex Tier

The Flex Tier offers a 50% discount on synchronous (real-time) requests, subject to capacity availability. When capacity is not immediately available, the request is queued and waits for up to 5 minutes before the API returns HTTP status code 429. If the queue is already full when the request arrives, the API returns 429 immediately.

How to use

To enable the Flex Tier, include the service_tier: "flex" parameter in your request:

import openai

client = openai.OpenAI(
    api_key="your-api-key",
    base_url="https://chat.maritaca.ai/api",
)

response = client.chat.completions.create(
    model="sabiazinho-4",
    messages=[
        {"role": "user", "content": "What is the capital of Brazil?"},
    ],
    extra_body={"service_tier": "flex"},
)

print(response.choices[0].message.content)
print(f"Service tier: {response.service_tier}")  # "flex"

Pricing

Flex requests have the same 50% discount as the Batch API, but are processed in real time. See the full pricing table on the Pricing page.

Flex vs Batch API

	Flex	Batch API
Discount	50%	50%
Response	Real-time (synchronous)	Up to 24h (asynchronous)
Availability	Subject to capacity (queue up to 5 min, or immediate 429 if queue is full)	Guaranteed within 24h window
Streaming	Yes	No
Best for	Applications that tolerate retry	Non-urgent batch processing

Best practices

1. Combine with prompt caching: cached tokens keep the 75% discount on input price. The Flex discount (50%) applies only to non-cached input tokens and output tokens — the discounts do not multiply. See more details at Prompt Caching.
2. Implement retry with backoff: since Flex requests may return 429, add retry logic with exponential backoff.

import openai
import time
import random

client = openai.OpenAI(
    api_key="your-api-key",
    base_url="https://chat.maritaca.ai/api",
)

def chat_flex_with_retry(messages, model="sabiazinho-4", max_retries=5):
    for attempt in range(max_retries):
        try:
            response = client.chat.completions.create(
                model=model,
                messages=messages,
                extra_body={"service_tier": "flex"},
            )
            return response
        except openai.RateLimitError:
            if attempt == max_retries - 1:
                raise
            wait = (2 ** attempt) + random.random()  # jitter to avoid simultaneous retries
            print(f"429 received. Retrying in {wait:.1f}s...")
            time.sleep(wait)

response = chat_flex_with_retry(
    messages=[{"role": "user", "content": "What is the capital of Brazil?"}]
)
print(response.choices[0].message.content)