If you have sized a project against Groq's free tier recently, the number you probably wrote down was 14,400 requests per day. It appears in a lot of comparison posts. It is on Groq's own rate limits page too, which is why it keeps propagating.
It just isn't attached to a model you would chat with.
Groq's Free Plan limits are published per model. As of today the chat models sit at 30 RPM and 1,000 RPD: openai/gpt-oss-120b, openai/gpt-oss-20b, and qwen/qwen3.6-27b all get 30 requests per minute, 1,000 per day, 8K tokens per minute, 200K per day. groq/compound is lower still at 250 RPD.
The two rows carrying 14.4K RPD are meta-llama/llama-prompt-guard-2-22m and meta-llama/llama-prompt-guard-2-86m. Those are 22M and 86M parameter classifiers whose job is screening prompts for injection attempts. They are meant to be called on every inbound message, which is exactly why their ceiling is high. They do not generate text.
So the practical gap is about 14x. If you planned an agent loop assuming 14,400 generations a day on the free plan, the real budget is 1,000, and you will find out when the 429s start.
Worth checking your own console rather than trusting any table, including this one, since these move without announcement. The rate limits page shows a Free Plan tab and a Developer plan tab, and the per-model rows are the only figures that mean anything.
Fuller comparison of the three fast free APIs: https://toolfreebie.com/groq-vs-cerebras-vs-gemini/
Top comments (0)