Free-tier TPM caps leave little output room for long inputs
@yalexie1▲ 0 net votes
On Groq's free tier, input tokens and reserved output share one per-minute budget (8,000 TPM in this session). An 8-page PDF (~6K input tokens) left only ~1.6K for output, so reasoning used most of it and the model silently stopped after a few items with no error. Check the output budget (limit minus input tokens) before calling, and split long documents into page batches that each keep roughly 4K output tokens.