Locus product guide · Updated November 10, 2026

Groq API pricing in 2026: GPT-OSS, Qwen and Whisper rates

Groq charges $0.15 per million input tokens and $0.60 per million output tokens for GPT-OSS 120B, $0.075 and $0.30 for GPT-OSS 20B, and $0.80 and $4.00 for Qwen 3.8 27B, a preview model. Whisper Large V3 transcription costs $0.111 per hour of audio, or $0.04 for the Turbo version. Batch jobs cost 50% less, cached input costs 50% less, and Llama models are now Enterprise only. Prices checked November 10, 2026.

How much does each Groq model cost?

Groq lists prices per million tokens on its models page, next to speed and rate limits. Production models are meant for live traffic. Preview models are for evaluation and can be removed at short notice. Prices checked November 10, 2026.

Meta Llama 3.1 8B, Llama 3.3 70B, MiniMax M2.7 and Muse Spark Open now show Contact Sales and an Enterprise label. Self-serve plans no longer list a price for them.

How much does each Groq model cost?
Model	Status	Input per 1M	Output per 1M	Speed	Context window
openai/gpt-oss-120b	Production	$0.15	$0.60	About 500 tokens a second	131,072
openai/gpt-oss-20b	Production	$0.075	$0.30	About 1,000 tokens a second	131,072
qwen/qwen3.8-27b	Preview	$0.80	$4.00	About 450 tokens a second	131,072
openai/gpt-oss-safeguard-20b	Preview	$0.075	$0.30	About 1,000 tokens a second	131,072

What do Groq Whisper transcription and speech cost?

Groq bills speech to text per hour of audio. Files can be up to 100 MB. Groq also lists Canopy Labs Orpheus text to speech at $22 per million characters for English and $40 for Saudi Arabic, both in preview.

What do Groq Whisper transcription and speech cost?
Model	Price	Cost of 100 hours	Developer plan limits
whisper-large-v3	$0.111 per hour	$11.10	200K audio seconds an hour, 300 requests a minute
whisper-large-v3-turbo	$0.04 per hour	$4.00	400K audio seconds an hour, 400 requests a minute

Deepgram vs Whisper

What does the Groq free plan include?

Groq has a Free plan with low limits and a Developer plan that pays per token. Limits apply to the whole organization, and you hit whichever comes first. Cached tokens do not count toward limits. The Developer plan also unlocks Batch and Flex processing.

What does the Groq free plan include?
Model	Free plan	Developer plan
openai/gpt-oss-120b	30 requests a minute, 1,000 a day, 8K tokens a minute, 200K a day	1,000 requests a minute, 250K tokens a minute
qwen/qwen3.8-27b	30 requests a minute, 1,000 a day, 8K tokens a minute	1,000 requests a minute, 250K tokens a minute
whisper-large-v3-turbo	20 requests a minute, 2,000 a day, 7,200 audio seconds an hour	400 requests a minute, 400K audio seconds an hour

How do Batch and prompt caching lower Groq costs?

Both discounts take 50% off, but they do not stack. Batch tokens bill at the batch rate whether or not they hit the cache.

Batch API: 50% off synchronous prices, with a processing window you set from 24 hours to 7 days. Batches use separate rate limits. If a batch expires, you pay only for completed requests.
Prompt caching: automatic on GPT-OSS 120B, GPT-OSS 20B and GPT-OSS Safeguard 20B, with no extra fee. Cached input tokens cost 50% less. Groq does not guarantee a cache hit.

What does a real Groq workload cost?

Take 10,000 calls, each with a 2,000-token prompt and a 500-token answer: 20 million input tokens and 5 million output tokens. The cached column assumes 1,500 tokens of each prompt hit the cache. Batch halves the standard column on models Batch supports.

One GPT-OSS 120B call of this size costs $0.0006. Speed is the other reason teams pick Groq: GPT-OSS 20B runs at about 1,000 tokens a second.

What does a real Groq workload cost?
Model	Standard	With 1,500 tokens cached	Cost per call
openai/gpt-oss-20b	$3.00	$2.44	$0.0003
openai/gpt-oss-120b	$6.00	$4.88	$0.0006
qwen/qwen3.8-27b	$36.00	No caching discount	$0.0036

How does Groq work through Locus Pro?

Locus Pro offers Groq chat as the groq/chat endpoint, paid from one prepaid credit balance with no Groq account. It accepts GPT-OSS 20B, GPT-OSS 120B and Qwen 3.8 27B in the OpenAI chat format. Catalog checked November 10, 2026.

Groq Whisper and the Batch API are not in the Locus catalog. For a high volume of small calls, Groq can cost less per call with your own Groq account and API key.

Locus Pro fits an agent that uses a fast open model for cheap steps and Claude or Gemini for hard ones. Every call is paid from one balance, shows its price first and can carry a ceiling, and a monthly limit stops new calls when you reach it.

When is Groq the right choice?

Pick Groq when latency matters and an open model is good enough: classification, extraction, routing and fast drafts. For high volume, batch whatever can wait and reuse prompt prefixes so they can hit the cache.

Cheapest text model: GPT-OSS 20B at $0.075 and $0.30.
Better reasoning at low cost: GPT-OSS 120B at $0.15 and $0.60.
File input: Qwen 3.8 27B, the only chat model with a listed file limit (20 MB).
Cheapest transcription: Whisper Large V3 Turbo at $0.04 an hour.

Frequently asked questions

Is Groq the same as Grok?

No. Groq is an inference company that serves open models on its own chips. Grok is the model family from xAI, with separate pricing.

Is the Groq API free?

There is a Free plan with low limits, such as 30 requests a minute and 8,000 tokens a minute on GPT-OSS 120B. The Developer plan pays per token and raises limits.

Can I still run Llama models on Groq?

Llama 3.1 8B and Llama 3.3 70B now show Enterprise and Contact Sales on the models page, with no self-serve price.

Do Groq Batch and caching discounts stack?

No. Each takes 50% off, but batch tokens bill at the batch rate whether or not they hit the cache.

Implementation references

Use these first-party references for current request contracts and account requirements. Tool availability, prices, and negotiated terms can change.

Sources

Third-party details were checked on November 10, 2026. Vendors change pricing and features often, so confirm on their site before you decide.