Groq Pricing Explained: Plans, Costs and What You Get
Groq pricing explained: free tier limits, per-token rates for GPT-OSS 120B, GPT-OSS 20B and Qwen, Batch API discounts, and how to estimate your monthly bill.
By Galactic News Editorial · Published 2026-11-07T22:38:15+00:00 · Updated 2026-11-07T22:38:15+00:00
Groq Pricing Explained: Plans, Costs and What You Get
Groq charges by the token. There is no flat monthly subscription for API access. You pay per million tokens, split into an input rate and a higher output rate. A free tier covers prototyping. This guide explains Groq pricing as listed on Groq's own documentation.
Groq pricing: the three tiers
Groq has three tiers: Free, Developer, and Enterprise. The Free tier costs $0 and needs no credit card. It is built for prototyping and evaluation, with tight per-model rate limits. The Developer tier is pay-as-you-go: add a payment method and pay per token. Enterprise is quote-based, for teams that need higher limits, support, and access to models that are not on the self-serve list.
The free tier limits vary by model. On GPT-OSS 120B, the free tier allows 30 requests a minute and 1,000 requests a day, with 8,000 tokens a minute and 200,000 tokens a day. The smaller models get higher daily allowances. Run past a limit and the API returns a 429 error.
Groq pricing per model
Groq runs open-weight models from other labs on its own LPU hardware. Prices are set by Groq per model. The table below shows the self-serve production models listed on Groq's documentation page, with prices per million tokens.
| Model | Input per 1M tokens | Output per 1M tokens | Speed |
|---|---|---|---|
| GPT-OSS 120B | $0.15 | $0.60 | ~500 tokens/sec |
| GPT-OSS 20B | $0.075 | $0.30 | ~1,000 tokens/sec |
| GPT-OSS Safeguard 20B | $0.075 | $0.30 | ~1,000 tokens/sec |
| Qwen3.8-27B | $0.80 | $4.00 | ~450 tokens/sec |
| Llama Prompt Guard 2 22M | $0.03 | $0.03 | Guardrail model |
| Llama Prompt Guard 2 86M | $0.04 | $0.04 | Guardrail model |
Two things stand out. GPT-OSS 20B is the cheapest general model, at $0.075 input and $0.30 output per million tokens. Qwen3.8-27B is the most expensive listed model, at $0.80 input and $4.00 output. Cached input costs half price on the GPT-OSS models.
Speech is billed differently. Whisper Large V3 costs $0.111 per hour of audio. Whisper Large V3 Turbo costs $0.04 per hour. Both models accept files up to 100 MB.
Note that the Llama 3.1 8B and Llama 3.3 70B models are now Enterprise-only. They show as Contact Sales on the models page. Older pricing guides still list them at $0.05 and $0.59 per million input tokens. Those rates no longer apply to self-serve accounts.
How to estimate your Groq bill
Work out your monthly tokens first. A million tokens is roughly 750,000 words of English text. A chatbot that handles 10,000 short conversations a month, at about 1,000 input tokens and 300 output tokens each, uses around 10 million input tokens and 3 million output tokens.
On GPT-OSS 120B, that workload costs about $1.50 in input and $1.80 in output, or $3.30 a month. On GPT-OSS 20B, the same workload costs about $0.75 in input and $0.90 in output, or $1.65 a month. These are rough estimates. Your real bill depends on your prompt sizes and response lengths.
Two discounts cut the bill further. The Batch API, for non-urgent asynchronous work, cuts the standard token rate in half. Prompt caching cuts cached input tokens to half price. Both can apply together.
Groq pricing vs the alternatives
Groq's main selling point is speed, not the lowest price. Its LPU hardware delivers hundreds of tokens per second on large models. That speed matters for real-time chat, voice agents, and any product where users wait for the reply.
For raw price, smaller models on Groq are cheap. GPT-OSS 20B at $0.075 per million input tokens is hard to beat for simple tasks. For heavy reasoning or long outputs, the output rate matters more than the input rate. Qwen3.8-27B at $4.00 per million output tokens can get expensive fast on long responses.
Check the live rate on Groq's own models page before you budget a production workload. Groq changes its model list often. Preview models can be discontinued at short notice, so build on production models only.
Getting started with Groq
Getting started takes minutes. Create an account at console.groq.com and generate an API key. No credit card is needed for the free tier. The API is OpenAI-compatible, so existing code that calls the OpenAI API can switch endpoints with minimal changes.
Start with GPT-OSS 20B for simple tasks and GPT-OSS 120B for harder ones. Measure your token use for a week before you add a payment method. The free tier limits will show up fast if your workload is real. That is the signal to move to the Developer tier.
Watch the rate limits as you scale. The Developer tier raises them well above the free tier. On GPT-OSS 120B, Developer allows 1,000 requests a minute and 250,000 tokens a minute. If you need more, Enterprise is the next step.
Bottom line
Groq pricing is simple: a free tier for testing, pay-per-token Developer pricing from $0.075 per million input tokens, and quote-based Enterprise. GPT-OSS 20B is the value pick. GPT-OSS 120B is the general-purpose default. Start on the free tier, measure your real token use, and upgrade when rate limits, not cost, become the problem.