AI / Claude Models Basics Interview Questions
What are Claude's rate limits and how are they structured?
Claude API rate limits prevent overload and ensure fair access. They are applied at three levels: requests per minute (RPM), tokens per minute (TPM), and tokens per day (TPD). Limits vary by model and by API usage tier.
| Limit type | What it restricts |
|---|---|
| Requests per minute (RPM) | Number of API calls per minute |
| Tokens per minute (TPM) | Total input + output tokens processed per minute |
| Tokens per day (TPD) | Total tokens processed in a 24-hour period |
Usage tiers: accounts start at Tier 1 with conservative limits and automatically advance to higher tiers as they spend more on the API (e.g. Tier 2 after $50 spend, Tier 3 after $500, Tier 4 after $5,000, Tier 5 after $50,000). Higher tiers get higher rate limits.
When rate limits are hit:
- The API returns a
429 RateLimitErrorresponse - Implement exponential backoff with jitter when retrying
- The Anthropic Python and TypeScript SDKs handle retries automatically by default (up to 2 retries)
- Rate limits can be increased by contacting Anthropic for approved use cases
Rate limits for models on Amazon Bedrock and Google Cloud are governed by those platforms separately and may differ from direct API limits.
What HTTP status code does the Claude API return when a rate limit is exceeded?
How does an API account advance from Tier 1 to higher rate limit tiers?
More Related questions...
