# Groq API pricing and models

Groq publishes per-token prices for 4 text models, as of October 20, 2026. The lowest output rate is $0.30 per million tokens, on GPT OSS 20B, and the highest-scoring model it serves is Qwen3.8-27B, in preview, at 73 on our 100-point score.

Groq serves open-weight models on its own LPU chips and sells them per token through GroqCloud.

Source: [GroqDocs supported models](https://console.groq.com/docs/models), read 2026-10-20. Prices in US dollars per million tokens.

## Models we score

| Model | Score | Input | Output | Cached input | Context | Listed speed | vs maker |
|---|---|---|---|---|---|---|---|
| Qwen 3.8 27B (`qwen/qwen3.8-27b`) | 73 | $0.80 | $4.00 | — | 131K | 450 tok/s | — |
| GPT OSS 120B (`openai/gpt-oss-120b`) | 55 | $0.15 | $0.60 | $0.075 | 131K | 500 tok/s | — |
| GPT OSS 20B (`openai/gpt-oss-20b`) | 50 | $0.075 | $0.30 | $0.037 | 131K | 1000 tok/s | — |

## Other models

| Model | Input | Output | Cached input | Context | Listed speed |
|---|---|---|---|---|---|
| Safety GPT OSS 20B (`openai/gpt-oss-safeguard-20b`) | $0.075 | $0.30 | $0.037 | 131K | 1000 tok/s |

## Billing notes

- Groq’s pricing page carries no rate table. The per-token prices sit on each model’s page in the GroqDocs, which is where we read them.
- Llama 3.1 8B, Llama 3.3 70B, Muse Spark Open and MiniMax M2.7 are listed for enterprise customers with no public price, so they are not in the table.

## What we measured (2026-09-01)

Groq hosts neither reference model, so it receives no score. On qwen3.8-27b it returned a 0.43-second median at $0.89 per million tokens blended, more per token than DeepInfra charged for GLM-5.2, and it was the only platform to return rate-limit errors mid-run (250,000 tokens per minute on the on-demand tier).

| Model run | Invoiced /1M blended | Median latency | p95 latency | Succeeded |
|---|---|---|---|---|
| qwen/qwen3.8-27b | $0.890 | 0.43 s | 2.00 s | 99.1% |

Method and invoices: https://benchlm.ai/openrouter-alternatives

## Hosts that sell the same models

- [DeepInfra](https://benchlm.ai/hosts/deepinfra): 3 shared, 3 cheaper per output token (GPT-OSS 120B: $0.17 vs $0.60)
- [OpenRouter](https://benchlm.ai/hosts/openrouter): 3 shared, 2 cheaper per output token (GPT-OSS 20B: $0.09 vs $0.30)
- [Novita AI](https://benchlm.ai/hosts/novita): 2 shared, 2 cheaper per output token (GPT-OSS 120B: $0.25 vs $0.60)
- [Cerebras](https://benchlm.ai/hosts/cerebras): 2 shared, 1 cheaper per output token (Qwen3.8-27B: $1.49 vs $4.00)
- [Nebius Token Factory](https://benchlm.ai/hosts/nebius): 2 shared, 1 cheaper per output token (Qwen3.8-27B: $3.00 vs $4.00)
- [Baseten](https://benchlm.ai/hosts/baseten): 1 shared, 1 cheaper per output token (GPT-OSS 120B: $0.50 vs $0.60)

## FAQ

### How much does the Groq API cost?

Groq bills per million tokens, with separate input and output rates for each model. On October 20, 2026 its output rates ran from $0.30 on GPT OSS 20B to $4.00 across 4 paid models. Input is usually cheaper than output, so a long prompt with a short reply costs less than the output rate suggests.

### Which models can you run on Groq?

4 text models are listed, and 3 of them match a model we score. The highest-scoring is Qwen3.8-27B at 73. The table on this page has every model with its input, output and cached-input rate, the context window, and the date we read the price.

### Is Groq fast?

In our paid September 1, 2026 run, Groq returned a median of 0.43 seconds and a p95 of 2.00 seconds on qwen/qwen3.8-27b, over 450 requests. Its docs list 450 to 1,000 tokens per second, depending on the model.

### What are the alternatives to Groq?

DeepInfra lists 3 of the same models and charges less output per token on 3 of them. OpenRouter and Novita AI also overlap. Price is only part of the choice: speed, rate limits and which model snapshot a host serves differ too.

Canonical page: https://benchlm.ai/hosts/groq