# Cerebras: Wafer-Scale Speed

Cerebras runs inference on wafer-scale chips, achieving speeds that seem impossible. 2000+ tok/s for Llama 3.1.

## Why Cerebras?

- **Blazing Speed** — 2000+ tokens/second
- **Open Models** — Llama 3.1, Mistral
- **Free Tier** — Generous rate limits
- **Real-Time Feel** — Instant responses

## Setup

```json
{
  "provider": "cerebras",
  "model": "llama3.1-70b"
}
```

## Try Ishi Today

Download Ishi and start automating your workflow with the Glass Box philosophy.