What it is
Cerebras serves open models on its own Wafer-Scale Engine chips instead of GPUs, through an OpenAI-compatible API and a developer console at cloud.cerebras.ai. Its public models currently include OpenAI's gpt-oss-120b, which its docs list at about 3,000 tokens per second, and Qwen 3.8 27B. It is also reachable through OpenRouter, Hugging Face, AWS Marketplace and Vercel.
It suits developers who need fast responses from open models. New accounts get a one-time $5 credit that expires after 30 days; after that the Developer tier is pay per token, topped up from $10, with higher rate limits, and Enterprise adds dedicated queue priority and custom model weights. The model list is very short, so check that it carries the model you need.
Best for
- Agents and coding tools where tokens per second matter
- Running gpt-oss and Qwen models through an API
- Comparing inference speed against GPU providers