Pricing snapshot
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Decision point | What Groq states |
|---|---|
Language models | Price per 1M tokens, with separate input and output rates |
Speech to text and text to speech | Whisper models are priced per hour; Orpheus models per 1M characters |
Rate limits | Listed per model for the Developer plan: TPM and RPM for text and speech-generation models, ASH and RPM for Whisper |
Data centers | 13 data centers across four continents, in the USA, Canada, Finland, the UK, Australia and Saudi Arabia |
Which model group fits your workload
Production models for live traffic. Groq says production models are intended for use in production environments. GPT OSS 120B, GPT OSS 20B and the two Whisper models are in this group with published rates.
Preview models for evaluation. Groq says preview models are for evaluation only, should not be used in production and may be discontinued at short notice. Qwen3.8 27B, Safety GPT OSS 20B, the two Prompt Guard models, the two Orpheus models, and the Enterprise models Muse Spark Open and MiniMax M2.7 are in this group.
Dedicated capacity. Groq's platform has three layers, and each includes the layers below: GroqMetal for dedicated bare-metal infrastructure under your control, GroqCore to add a tested inference stack with no infrastructure expertise required, and GroqAssured to add governance, auditability and control.
Published model rates
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Model (group) | Price | Rate limits (Developer plan) and context |
|---|---|---|
GPT OSS 120B (Production) | $0.15 input, $0.60 output per 1M tokens | 250K TPM, 1K RPM; 131,072-token context, 65,536 max completion tokens |
GPT OSS 20B (Production) | $0.075 input, $0.30 output per 1M tokens | 250K TPM, 1K RPM; 131,072-token context, 65,536 max completion tokens |
Whisper Large V3 (Production) | $0.111 per hour | 200K ASH, 300 RPM; 100 MB max file size |
Whisper Large V3 Turbo (Production) | $0.04 per hour | 400K ASH, 400 RPM; 100 MB max file size |
Qwen3.8 27B (Preview) | $0.80 input, $4.00 output per 1M tokens | 250K TPM, 1K RPM; 131,072-token context, 16,384 max completion tokens, 20 MB max file size |
Safety GPT OSS 20B (Preview) | $0.075 input, $0.30 output per 1M tokens | 150K TPM, 1K RPM; 131,072-token context |
Llama Prompt Guard 2 22M / Prompt Guard 2 86M (Preview) | 22M: $0.03 input, $0.03 output; 86M: $0.04 input, $0.04 output per 1M tokens | 30K TPM, 100 RPM; 512-token context |
Orpheus V1 English / Orpheus Arabic Saudi (Preview) | English: $22.00 per 1M characters; Arabic Saudi: $40.00 per 1M characters | 50K TPM, 250 RPM each |
Llama 3.1 8B, Llama 3.3 70B (Production, Enterprise) | Contact Sales | Rate limits: Contact Sales; 131,072-token context |
Muse Spark Open, MiniMax M2.7 (Preview, Enterprise) | Contact Sales | Rate limits: Contact Sales; context 131,072 (Muse Spark Open) and 196,608 (MiniMax M2.7) tokens |
How the bill is built
For a language model, the charge is input tokens times the model's input rate plus output tokens times its output rate. Because each model has its own pair of rates, a request mix spread over several models is priced model by model.
Worked examples
These are hypothetical monthly volumes worked from the published rates above. Each covers model usage only.
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Assumed usage | Working (millions of tokens or hours x rate) and result |
|---|---|
GPT OSS 120B: 10M input and 2M output tokens | 10 x $0.15 = $1.50 input + 2 x $0.60 = $1.20 output = $2.70 per month, tokens only |
GPT OSS 120B: 200M input and 40M output tokens | 200 x $0.15 = $30 input + 40 x $0.60 = $24 output = $54 per month, tokens only |
GPT OSS 120B: 1,000M input and 200M output tokens | 1,000 x $0.15 = $150 input + 200 x $0.60 = $120 output = $270 per month, tokens only |
GPT OSS 20B: 1,000M input and 200M output tokens | 1,000 x $0.075 = $75 input + 200 x $0.30 = $60 output = $135 per month, tokens only |
Whisper Large V3 Turbo: 100 hours of audio | 100 x $0.04 per hour = $4, transcription only |
What drives the bill
Output tokens. GPT OSS 120B, GPT OSS 20B, Safety GPT OSS 20B and Qwen3.8 27B each list a higher rate per 1M output tokens than per 1M input tokens, so on these models 1M output tokens costs more than 1M input tokens.
Model choice. Rates differ between most models, so the model you route a workload to sets its rate; GPT OSS 20B and Safety GPT OSS 20B share the same rates.
Costs that are easy to miss
Throughput ceilings. Developer plan limits differ by model, from 30K TPM and 100 RPM on the Prompt Guard models to 250K TPM and 1K RPM on GPT OSS 120B, so check a model's limit as well as its rate before you size a workload.
Completion caps. GPT OSS models list up to 65,536 max completion tokens and Qwen3.8 27B up to 16,384; plan long outputs around these caps.
Our take
If your workload runs on a production model with a published rate, you can price Groq with plain arithmetic before you build; if it needs Llama 3.1 8B or Llama 3.3 70B in production, get a quote before you commit. Muse Spark Open and MiniMax M2.7 also need a quote, but as preview models they suit evaluation, not production.