AI cost · Groq
Groq API cost: what you actually pay
Groq's speed is free, but the free tier's tiny rate limits (30 req/min, 14,400/day) push teams to add a card — which also happens to cut the price 25% — and the hosted-model roster rotates often enough that a pinned model ID can disappear.
Public pricing
Token rates (USD per 1M tokens)
| Model | Input | Output | Cache read |
|---|---|---|---|
| Llama 3.3 70B Versatile | $0.59 | $0.79 | — |
| Llama 3.1 8B Instantcheapest widely used model | $0.05 | $0.08 | — |
| Kimi K2 Instruct | $1 | $3 | $0.5 |
| GPT OSS 120B | $0.15 | $0.6 | $0.075 |
Token price is identical across On-Demand/Flex/Performance service tiers — tiers govern throughput and queueing, not $/token. Batch API and cached-input hits are each 50% off the on-demand input rate and stack (~25% of original price combined). Adding a card unlocks 'Developer Tier': a flat 25% discount plus 10x higher rate limits versus the free tier.
Last updated: 2026-07-22. Sourced from Groq Pricing (official) ↗, Groq Service Tiers Docs (official) ↗.
Where the real Groq API cost hides
The sticker is the smallest part of the story. For Groq API, these are the line items that quietly inflate the bill:
- Free tier is capped at 30 requests/min and 14,400/day — many teams hit this in testing, but the fix (adding a card for Developer Tier) is also a 25% price cut, so staying 'free' can mean an implicit premium in dev time and retries.
- Groq periodically rotates which open models it hosts (deprecating older Llama/Mixtral endpoints); a model ID that worked last quarter can disappear, forcing an unplanned re-pick at a different price point.
- Cached-input and batch discounts require specific request parameters/endpoints; default synchronous calls with fresh context every time get none of the available 50% discounts.
What an audit finds on Groq API
Reflexive use of a larger hosted model (e.g., Kimi K2 Instruct at $1/$3 per M) for simple classification/extraction tasks that Llama 3.1 8B Instant ($0.05/$0.08) would handle at roughly 1/15th the cost.
Groq API pricing FAQ
Does Groq's speed advantage cost extra?
No — per-token price is the same across service tiers; you only pay more if you upgrade to the Enterprise 'Performance' tier for guaranteed low-latency capacity.
Can I get a volume discount without a custom contract?
Yes — Groq stacks 50% off cached input tokens and 50% off Batch API jobs, plus a 25% Developer Tier discount for adding a payment method, all without an enterprise agreement.
What is Groq API really costing you?
Spendassay connects your Groq API spend to actual usage and engineering output — a proof-level AI cost report with the wasted dollars named. Neutral across every AI vendor.
Free · read-only · no card
Other api platforms pricing
More: the complete AI cost management guide · every tool, with prices · cost per outcome · compare platforms · how we measure