Groq API Pricing (2026)
Groq provides ultra-fast LLM inference using custom LPU hardware, delivering the fastest token generation speeds on the market for open-source models. Founded in 2016 and headquartered in Mountain View, CA, Groq offers 8 active models through their API. All pricing data below is sourced directly from the Groq pricing page and is updated regularly.
Prices verified Mar 6, 2026
Groq Model Pricing — All Models
All prices are in USD per 1 million tokens ($/M tokens). Input tokens are the text you send to the model; output tokens are the generated response.
| Model | Tier | Input | Output | Cached Input | Batch Input | Batch Output |
|---|---|---|---|---|---|---|
GPT-OSS 120B (Groq) A 120B parameter open-source GPT model running on Groq's LPU hardware for ultra-fast inference. | Mid | $0.15/M | $0.60/M | — | — | — |
Kimi K2 (Groq) Moonshot AI's Kimi K2 model running on Groq for fast inference on complex reasoning and coding tasks. | Premium | $1.00/M | $3.00/M | — | — | — |
Llama 4 Scout (Groq) Meta's compact Llama 4 Scout model on Groq, offering great value for lightweight tasks with ultra-fast inference. | Budget | $0.11/M | $0.34/M | — | — | — |
Llama 4 Maverick (Groq) Meta's Llama 4 Maverick on Groq hardware, balancing strong performance with fast, affordable inference. | Mid | $0.20/M | $0.60/M | — | — | — |
Qwen3 32B (Groq) Alibaba's Qwen3 32B model on Groq for fast inference on reasoning and general tasks. | Mid | $0.29/M | $0.59/M | — | — | — |
Llama 3.3 70B (Groq) Meta's Llama 3.3 70B running on Groq's LPU hardware, delivering extremely fast inference with low latency. | Mid | $0.59/M | $0.79/M | — | — | — |
Llama 3.1 8B (Groq) Meta's Llama 3.1 8B on Groq's LPU hardware for the fastest possible low-cost inference. | Budget | $0.05/M | $0.08/M | — | — | — |
Mixtral 8x7B (Groq)(deprecated) Mistral's Mixtral 8x7B mixture-of-experts model running on Groq, offering good performance at low cost with fast inference. | Budget | $0.24/M | $0.24/M | — | — | — |
GPT-OSS 20B (Groq) OpenAI's compact open-weight 20B MoE model running on Groq's LPU hardware for ultra-fast, cost-efficient inference. | Budget | $0.08/M | $0.30/M | $0.04/M | — | — |
Prices last verified: 2026-03-06. Always confirm on the official pricing page before production use.
Groq Model Capabilities
Key technical capabilities across all Groq models. Use this table to identify which models support the features your application requires.
| Model | Context | Max Output | Vision | Functions | JSON Mode | Caching | Batch API | Reasoning |
|---|---|---|---|---|---|---|---|---|
| GPT-OSS 120B (Groq) | 131,072 | 32,768 | — | ✓ | ✓ | — | — | — |
| Kimi K2 (Groq) | 256,000 | 32,768 | — | ✓ | ✓ | — | — | — |
| Llama 4 Scout (Groq) | 131,072 | 8,192 | ✓ | ✓ | ✓ | — | — | — |
| Llama 4 Maverick (Groq) | 131,072 | 8,192 | ✓ | ✓ | ✓ | — | — | — |
| Qwen3 32B (Groq) | 131,072 | 8,192 | — | ✓ | ✓ | — | — | ✓ |
| Llama 3.3 70B (Groq) | 131,072 | 32,768 | — | ✓ | ✓ | — | — | — |
| Llama 3.1 8B (Groq) | 131,072 | 8,000 | — | ✓ | ✓ | — | — | — |
| GPT-OSS 20B (Groq) | 131,072 | 32,768 | — | ✓ | ✓ | ✓ | — | — |
When to Choose Groq
Cost-sensitive workloads
Use Llama 3.1 8B (Groq) — Groq's most affordable model at $0.05/M input / $0.08/M output. Estimated cost at 100K requests/month: $9.00.
Complex or high-accuracy tasks
Use Kimi K2 (Groq) — Groq's most capable model (premium tier) with a 256,000-token context window. Priced at $1.00/M input / $3.00/M output.
High-volume pipelines
Groq models are well-suited for high-throughput workloads. Consider batch API (where supported) for an additional 50% cost reduction on asynchronous pipelines. Check the pricing table above for batch pricing availability.
Compare Groq Models vs Competitors
See detailed cost comparisons between Groq models and models from other providers, including side-by-side pricing tables and monthly cost projections.
Use Cases for Groq Models
Browse use case guides where Groq models are recommended, including cost-effective model rankings and monthly cost estimates for each workload.
Groq Free Tier Details
The following Groq models include a free tier for development and prototyping. Free tier limits apply to unpaid usage only.
| Model | Requests/min | Requests/day | Expires |
|---|---|---|---|
| GPT-OSS 120B (Groq) | 30 | 1,000 | No expiry |
| Kimi K2 (Groq) | 30 | 1,000 | No expiry |
| Llama 4 Scout (Groq) | 30 | 14,400 | No expiry |
| Llama 4 Maverick (Groq) | 30 | 14,400 | No expiry |
| Qwen3 32B (Groq) | 30 | 14,400 | No expiry |
| Llama 3.3 70B (Groq) | 30 | 1,000 | No expiry |
| Llama 3.1 8B (Groq) | 30 | 14,400 | No expiry |
| GPT-OSS 20B (Groq) | 30 | 14,400 | No expiry |
Deprecated Groq Models
The following Groq models are deprecated and should not be used in new projects. Historical pricing is shown for reference only.
- Mixtral 8x7B (Groq) — deprecated 2025-12-01
Frequently Asked Questions: Groq API Pricing
What is the cheapest Groq model?
The cheapest Groq model by input token price is Llama 3.1 8B (Groq), priced at $0.05/M for input tokens and $0.08/M for output tokens. It is best suited for high-volume, cost-sensitive workloads where speed and price matter most.
What is the most capable Groq model?
Kimi K2 (Groq) is Groq's most capable model, classified as a premium-tier model. It supports a context window of 256,000 tokens and is priced at $1.00/M input / $3.00/M output per million tokens.
How does Groq API pricing work?
Groq charges per token consumed — separately for input tokens (the text you send) and output tokens (the text the model generates). Prices are listed in USD per 1 million tokens ($/M). Groq was founded in 2016 and is headquartered in Mountain View, CA. All pricing shown here is sourced from the official Groq pricing page.
Does Groq offer batch API pricing?
Groq's current model lineup does not include batch API pricing at this time. For cost reduction on high-volume workloads, consider prompt caching (where supported) or running multiple concurrent requests with standard pricing.
Which Groq models support prompt caching?
1 Groq model support prompt caching: GPT-OSS 20B (Groq). Prompt caching stores repeated context (system prompts, knowledge base content) in memory and applies a discount to cached input tokens, making it highly effective for applications that reuse the same context across many requests.
Which Groq models support vision / image inputs?
2 Groq models support vision (image input): Llama 4 Scout (Groq), Llama 4 Maverick (Groq). These models can analyze images, charts, screenshots, and documents alongside text prompts.
Groq pricing data last verified: 2026-03-06. Prices may change — verify on the official Groq pricing page before production use.