Rate limits & when to reach for Groq

Groq's free tier has real constraints. Knowing them upfront saves you from hitting a wall mid-experiment. This card also shows you when Cerebras or OpenRouter is the better choice — the three providers share the same API shape, so switching is a one-line change.

1. 1 Use your OWN free Groq key — never the 32dots project key. The shared key was suspended after a silent batch job spent €19.71 overnight. Your personal key is free, instant, and safe to experiment with.
2. 2 Understand the limit structure. Free tier: ~30 requests/minute, ~14,400 requests/day ORG-wide. But each model has its own daily cap — llama-3.3-70b-versatile is capped at ~1,000 req/day; llama-3.1-8b-instant is much higher. Getting a second key does NOT double your quota — limits are per organisation, not per key.
3. 3 Choose the right provider for the job. Open the Inference Providers cheatsheet (linked above) and compare: - Groq — best for: fast iteration, Whisper STT, n8n pipelines with moderate volume. Watch: per-model daily caps. - Cerebras — best for: high-volume free usage (1M tokens/day, no credit card), fastest throughput on big models. Watch: ~8K context cap on free tier. - OpenRouter — best for: widest model selection (GPT-4o, Claude, Gemma — your key, your choice). Watch: no universal free tier; each model has its own pricing.
4. 4 Do the swap mentally. If your Groq code hits a rate limit, you change exactly 3 things to move to Cerebras: base_url → https://api.cerebras.ai/v1, api_key → your Cerebras key, model → a Cerebras model name (e.g. llama-3.3-70b). Your SDK call, error handling, and response parsing are unchanged.
5. 5 Check console.groq.com → Usage → Rate limits right now. Verify your personal key's per-model caps. Bookmark it — it updates as Groq revises limits.