HomeAll ToolsCategoriesGuides & StrategyAI ToolsAbout UsExplore All Tools
Rate Limits

Free LLM API Rate Limits Explained: How to Avoid HTTP 429 Errors

The most common obstacle when developing against free LLM APIs is the dreaded HTTP 429 Too Many Requests response. Understanding how inference providers measure rate limits—and how to architect queueing and retry strategies—is critical for smooth development.

Interactive AI Model Explorer

Want to calculate your exact parameters and operational expenses? Run the calculations locally inside your browser.

Launch AI Model Explorer

1. The Three Primary Dimensions of Rate Limits

Providers govern server load using three distinct mathematical bounds:

1. Requests Per Minute (RPM): The maximum number of HTTP calls permitted in a rolling 60-second window. Common free limits range from 15 to 30 RPM. 2. Tokens Per Minute (TPM): The cumulative sum of input prompt tokens and output completion tokens processed within 60 seconds. A single large prompt can trigger a TPM violation even if your RPM count is only 1. 3. Requests Per Day (RPD): The absolute quota of API calls allowed per 24-hour cycle. Resets usually occur at 00:00 UTC.

2. Provider Rate Limits Breakdown

ProviderRPMTPMRPDConcurrency Limit
Google AI Studio15 RPM1,000,000 TPM1,500 RPD2 concurrent
Groq Cloud (70B)30 RPM6,000 TPM14,400 RPD3 concurrent
Groq Cloud (8B)30 RPM30,000 TPM14,400 RPD5 concurrent
OpenRouter Free20 RPMModel Specific200 RPD1 concurrent
GitHub Models15 RPM150,000 TPM150 RPD1 concurrent
Mistral (Experiment)60 RPM500,000 TPMFair Use2 concurrent

3. Architectural Strategies to Never Hit 429 Errors

Implement these three engineering patterns in your code:

  • Exponential Backoff with Jitter: When a 429 occurs, parse the Retry-After header. If absent, wait 2^attempt + random(0, 1) seconds before re-attempting.
  • Client-Side Token Bucket: Use an in-memory or Redis-backed token bucket algorithm to throttle outgoing calls before they hit the provider's network boundary.
  • Multi-Provider Fallback Routing: Chain Groq -> Google AI Studio -> OpenRouter. If Groq returns a rate limit, the request automatically reroutes to Google.

Frequently Asked Questions

When do daily rate limit quotas reset?

Google AI Studio and Groq reset their daily request count at 00:00 UTC (Universal Coordinated Time).

Does prompt caching count toward TPM rate limits?

On providers that support context caching (like Google), cache-hit tokens typically do not count against standard TPM thresholds.