Two numbers on Groq's free plan interact badly: 1,000 requests/day and 200,000 tokens/day per chat model.
200,000 / 1,000 = 200 tokens per request — system prompt, context, and completion combined. Nothing real fits in that. A modest exchange runs 2,000 tokens, which means the token budget dies at ~100 requests, 10% of the advertised request cap.
The asymmetry is invisible at runtime: response headers report requests-remaining (x-ratelimit-remaining-requests) but the daily token budget isn't in the headers at all. Your 429 arrives while the request counter says you have 900 left.
Exception worth knowing: qwen/qwen3.8-27b gets 2M tokens/day — ten times its siblings — making it the only model on the free plan you can drive hard all day.
Current per-model table and what fits in each budget: https://toolfreebie.com/groq-fastest-free-ai-api/
Top comments (0)