What affects cost
- Selected model endpoint
- Input and output token volume
- Text, speech, transcription, or other service unit
- Retries and application-level agent loops
- Production rate limits and enterprise-only model requirements
LLM API
GroqCloud API pricing
GroqCloud API pricing lists per-model input and output token rates for hosted open models, with spend controls and separate rates for audio services.
Decision summary: Consider GroqCloud when its supported model catalog, token rates, service limits, and spend controls fit the production workload.
Pricing overview
Usage-based model inference priced per million tokens, plus separate per-character or per-audio-hour units for non-text services.
| Plan or model | Pricing posture | Decision note |
|---|---|---|
| GPT OSS 20B 128k | $0.075 / 1M input; $0.30 / 1M output | Official Groq on-demand text-model rate. |
| GPT OSS 120B 128k | $0.15 / 1M input; $0.60 / 1M output | Official Groq on-demand text-model rate. |
| Llama 4 Scout 128k | $0.11 / 1M input; $0.34 / 1M output | Official rate for the listed Llama 4 Scout endpoint. |
| Qwen3 32B 131k | $0.29 / 1M input; $0.59 / 1M output | Official rate for the listed Qwen3 32B endpoint. |
| Llama 3.3 70B Versatile 128k | $0.59 / 1M input; $0.79 / 1M output | Official Groq on-demand text-model rate. |
| Llama 3.1 8B Instant 128k | $0.05 / 1M input; $0.08 / 1M output | Lowest tracked text-model rate in this StackLens snapshot; model fit still requires testing. |
Lower-cost options from the same provider
- Llama 3.1 8B Instant for suitable lower-complexity work
- GPT OSS 20B when its task quality is sufficient
- Cap generated output and monitor retries for every endpoint
Alternative providers or products
- Together AI
- OpenRouter
- Direct model-provider APIs
- Self-hosted inference when operational cost is understood
Best for
- teams evaluating Groq-hosted open models
- applications that need explicit per-model token rates
- buyers that want dashboard usage and spend controls
Not ideal for
- teams requiring models outside the current Groq catalog
- budgets that mix text and audio billing units
- workloads that have not tested endpoint availability and rate limits
FAQ
Questions teams ask before choosing
Are all GroqCloud models billed at the same token rate?
No. Each listed model has its own input and output rate, while speech and transcription services use different billing units.
Does the lowest token rate identify the right Groq model?
No. Token cost should be evaluated with task quality, output length, retries, context limits, endpoint availability, and rate limits.