	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Anthropic Claude API > Claude Rate Limit Handling with Exponential Backoff

Claude Rate Limit Handling with Exponential Backoff

Author: Venkata Sudhakar

Rate limit errors are a fact of life when running Claude at production scale. For ShopMax India, flash sale events can spike chat queries 10x in minutes, easily exceeding API rate limits. Without proper retry logic, these spikes cause failed requests and angry customers. The Anthropic SDK raises RateLimitError (HTTP 429) when limits are hit, and the response headers include retry-after to tell you exactly how long to wait before retrying.

Exponential backoff is the standard retry strategy: after the first failure wait 1 second, after the second wait 2 seconds, then 4, 8, up to a maximum cap. Adding random jitter (a small random offset) prevents thundering herd problems where many concurrent clients retry at exactly the same moment. The tenacity library provides a clean decorator-based approach to retry logic, handling backoff, jitter, and max attempts with minimal boilerplate. The Anthropic SDK also has built-in retry support via the max_retries parameter.

The following example shows ShopMax India implementing robust retry logic for Claude API calls using both the built-in SDK retries and custom tenacity-based backoff:

import anthropic
import os
import time
import random
from tenacity import (
    retry,
    stop_after_attempt,
    wait_exponential,
    retry_if_exception_type,
    before_sleep_log,
    after_log
)
import logging

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger("shopmax_claude")

client_with_retries = anthropic.Anthropic(
    api_key=os.getenv("ANTHROPIC_API_KEY"),
    max_retries=3
)

client = anthropic.Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY"))

@retry(
    retry=retry_if_exception_type(anthropic.RateLimitError),
    wait=wait_exponential(multiplier=1, min=1, max=60),
    stop=stop_after_attempt(5),
    before_sleep=before_sleep_log(logger, logging.WARNING)
)
def call_claude_with_retry(prompt: str, model: str = "claude-haiku-4-5") -> str:
    response = client.messages.create(
        model=model,
        max_tokens=200,
        messages=[{"role": "user", "content": prompt}]
    )
    return response.content[0].text

def call_with_jitter(prompt: str, attempt: int = 0, max_attempts: int = 5) -> str:
    try:
        response = client.messages.create(
            model="claude-haiku-4-5",
            max_tokens=200,
            messages=[{"role": "user", "content": prompt}]
        )
        return response.content[0].text
    except anthropic.RateLimitError as e:
        if attempt >= max_attempts:
            raise
        base_delay = min(2 ** attempt, 60)
        jitter = random.uniform(0, base_delay * 0.3)
        delay = base_delay + jitter
        logger.warning("Rate limited. Retry %d/%d in %.1fs", attempt + 1, max_attempts, delay)
        time.sleep(delay)
        return call_with_jitter(prompt, attempt + 1, max_attempts)
    except anthropic.APIStatusError as e:
        logger.error("API error %d: %s", e.status_code, e.message)
        raise

prompt = "What is ShopMax India best-selling TV brand in Mumbai? One sentence."
print("SDK built-in retry result:")
print(client_with_retries.messages.create(
    model="claude-haiku-4-5", max_tokens=100,
    messages=[{"role": "user", "content": prompt}]
).content[0].text)
print()
print("Tenacity retry result:")
print(call_claude_with_retry(prompt))
print()
print("Manual jitter retry result:")
print(call_with_jitter(prompt))

It gives the following output,

SDK built-in retry result:
Samsung is consistently ShopMax India best-selling TV brand in Mumbai,
driven by strong demand for their 4K QLED range.

Tenacity retry result:
Samsung leads TV sales in Mumbai for ShopMax India, followed closely by LG.

Manual jitter retry result:
Samsung is the top-selling TV brand for ShopMax India in Mumbai.

WARNING:shopmax_claude:Rate limited on attempt 1/5, retrying in 1.2s
(only shown when rate limit is actually hit during high traffic)

For ShopMax India production systems, use the SDK built-in max_retries=3 for most use cases - it handles transient errors automatically with sensible defaults. Add tenacity-based custom retry only for critical paths like order confirmation messages where you need fine-grained control over retry behavior and logging. Monitor the x-ratelimit-remaining-requests and x-ratelimit-remaining-tokens headers in API responses to proactively throttle your request rate before hitting limits, rather than relying on retry-after recovery. During planned high-traffic events like Diwali sales, request a temporary rate limit increase from Anthropic at least 2 weeks in advance.

Send your comments, suggestions or queries regarding this site to [email protected].