Groq API Pricing, Models & Rate Limits (2026)
Groq doesn't train models — it serves open-weight models (OpenAI's gpt-oss and Alibaba's Qwen among them) on its own LPU inference hardware. The pitch is raw throughput: Groq is consistently among the fastest tokens-per-second on this site's speed benchmarks for the weights it hosts.
Models and prices
| Model | Input /M | Output /M | Blended /M |
|---|---|---|---|
| GPT-OSS 20B | $0.07 | $0.30 | $0.13 |
| GPT-OSS 120B | $0.15 | $0.60 | $0.26 |
| Qwen 3.8 30B | $0.60 | $3.00 | $1.20 |
4 legacy Groq models
| Llama 3.1 8B | $0.06/M blended |
| Llama 4 Maverick | $0.30/M blended |
| Llama 3.3 70B | $0.64/M blended |
| Qwen 3.6 27B | $1.20/M blended |
Speed
Fastest measured Groq model is GPT-OSS 20B at 1120 tokens/sec (140ms TTFT), median across measured Groq models is 780 tokens/sec. See the full speed benchmark methodology.
Operational facts
Verified 2026-08-08 against source.
| OpenAI-compatible | Yes |
| API base URL | https://api.groq.com/openai/v1 |
| Auth model | Bearer API key |
| Prompt caching | Not documented |
| Batch discount | 50% |
| Free tier | Free tier with per-minute and per-day token caps |
| Rate-limit model | Per-model requests/minute and tokens/minute caps by tier |
| Data residency | US |
| Trains on API data | Not documented |
| SLA published | No |
Lifecycle
Groq has 4 legacy models still routable and 2 retired models. Next scheduled retirement: Llama 3.3 70B on 2026-08-16. Full dates and successors on the model deprecation tracker.
Switching to and from Groq
Calling Groq through All AI Ask
curl https://allaiask.com/api/v1/chat \
-H "Authorization: Bearer $ALLAIASK_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "gpt-oss-20b", "messages": [{"role": "user", "content": "Hello"}]}'FAQ
Is Groq OpenAI-compatible?
Yes — Groq's API base (https://api.groq.com/openai/v1) accepts the OpenAI SDK request/response shape, so existing OpenAI client code works with a base-URL and key swap.
Does Groq support prompt caching?
Not documented as of 2026-08-08 — we did not find a published prompt-caching feature for Groq. If that changes, this page updates.
Does Groq have a free tier?
Yes — Free tier with per-minute and per-day token caps.
How much does the Groq API cost?
Current Groq models range from $0.13 to $1.20 per million blended tokens (3:1 input:output). Full per-model pricing is in the table below.
Where is Groq API data hosted?
US