CS

Cerebras WSE-3 Turbo (WSE-3T)

Hardware

by Cerebras Systems

Wafer-scale processor — now WSE-3 Turbo (WSE-3T), four trillion transistors and 900,000 AI cores — sold inside Cerebras' rack-scale systems: three WSE-3T processors power the CS-4, the successor to the WSE-3-based CS-3.

Find alternatives to Cerebras WSE-3 Turbo (WSE-3T)

The published pricing page covers the inference API only; the hardware is quoted.

What it does

• Accelerator silicon
• Model inference
• Model training

How it compares

• NVIDIA Blackwell

  Remarkably, the Cerebras Wafer Scale Engine 3, launched in 2024, still outperforms NVIDIA’s latest Blackwell generation by almost 5x—underscoring the enduring advantage of a compute architecture purpose-built for large-scale AI inference.

  Official documentation · 20 Sep 2026
• SambaNova RDU

  Cerebras uses wafer-scale processors to deliver high-speed inference through its cloud, dedicated systems, and on-premise deployments.

  Research report · 19 Sep 2026

Sources

Pricing

Official documentation · 21 Aug 2026

The published pricing page covers the inference API only; the hardware is quoted.

What it does

Official documentation · 7 Sep 2026

Fetched 2026-09-07: HTTP 200, <title> “Product - Chip - Cerebras”. The page says “Train deep learning models faster, with lower power consumption and industry-leading efficiency” and carries a “Breakthrough Inference Performance” section. Both uses stated on the vendor's own page.

Sold within

Official documentation · 30 Sep 2026

"The Cerebras CS-4 rack-scale AI system is powered by three WSE-3T processors". The chip page names no way to buy the processor on its own.

Also from Cerebras Systems

• Cerebras Inference Model API

  Hosted inference service that serves open-weight models on Cerebras wafer-scale hardware, billed per token.

• Cerebras Training Cloud Platform

  Managed training/fine-tuning cloud for models from 1B to tens-of-trillions of parameters without manual sharding or model-parallelism configuration, billed pay-per-hour or pay-per-model (Cerebras-run custom training).

Open Cerebras Systems in the directory

Something wrong here? Send a correction — quote this product id: cerebras-wse.