Pricing / Groq

Groq API pricing

37 model entries. Groq-hosted entries in the bundled pricing snapshot. Source rate fields are shown as encoded in the snapshot dated 2026-03-25; model modality is not encoded.

Lowest encoded input rate

$0.04/1M

llama-3.2-1b-preview

Highest encoded input rate

$1/1M

kimi-k2-instruct-0905

Models with cache pricing

7 of 37

Interpret the /1M columns as token rates only for models independently verified as token-billed. Non-text billing units are not represented in this snapshot schema.

ModelInput $/1MOutput $/1MCache $/1M
llama-3.2-1b-preview$0.04$0.04-
llama-3.1-8b-instant$0.05$0.08-
llama3-8b-8192$0.05$0.08-
llama-3.2-3b-preview$0.06$0.06-
gemma-7b-it$0.07$0.07-
openai/gpt-oss-20b$0.075$0.3$0.0375
gpt-oss-20b$0.075$0.3$0.0375
gpt-oss-safeguard-20b$0.075$0.3$0.037
meta-llama/llama-4-scout-17b-16e-instruct$0.11$0.34-
llama-4-scout-17b-16e-instruct$0.11$0.34-
openai/gpt-oss-120b$0.15$0.6$0.075
gpt-oss-120b$0.15$0.6$0.075
llama-3.2-11b-text-preview$0.18$0.18-
llama-3.2-11b-vision-preview$0.18$0.18-
llama3-groq-8b-8192-tool-use-preview$0.19$0.19-
gemma2-9b-it$0.2$0.2-
llama-guard-3-8b$0.2$0.2-
meta-llama/llama-4-maverick-17b-128e-instruct$0.2$0.6-
meta-llama/llama-guard-4-12b$0.2$0.2-
llama-guard-4-12b$0.2$0.2-
llama-4-maverick-17b-128e-instruct$0.2$0.6-
mixtral-8x7b-32768$0.24$0.24-
qwen/qwen3-32b$0.29$0.59-
qwen3-32b$0.29$0.59-
llama-3.3-70b-versatile$0.59$0.79-
llama-3.1-405b-reasoning$0.59$0.79-
llama-3.1-70b-versatile$0.59$0.79-
llama-3.3-70b-specdec$0.59$0.99-
llama3-70b-8192$0.59$0.79-
llama2-70b-4096$0.7$0.8-
deepseek-r1-distill-llama-70b$0.75$0.99-
mistral-saba-24b$0.79$0.79-
llama3-groq-70b-8192-tool-use-preview$0.89$0.89-
llama-3.2-90b-text-preview$0.9$0.9-
llama-3.2-90b-vision-preview$0.9$0.9-
moonshotai/kimi-k2-instruct$1$3$0.5
kimi-k2-instruct-0905$1$3$0.5

Use the Groq snapshot as a reference, then measure the request

Local tracking can estimate supported responses from provider usage fields. The authenticated gateway path adds request identity, persisted cost evidence, session attribution, and pre-dispatch budget checks.